gavindiaz/NexQuant
1
0
Fork 0
mirror of https://github.com/NicolasBohn/NexQuant.git synced 2026-07-27 15:37:44 +00:00
Code Issues Packages Projects Releases Wiki Activity

chore: initial Predix state (RD-Agent fork + EURUSD setup)

Browse Source
This commit is contained in:
TPTBusiness
2026-03-22 21:20:02 +01:00
parent ffb9491c47
commit 30c0a9166e
51 changed files with 1026 additions and 8539 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.bumpversion.cfg
-6
View File
@@ -1,6 +0,0 @@
[bumpversion]
current_version = 0.0.0
commit = True
tag = True
[bumpversion:file:pyproject.toml]
.commitlintrc.js
-21
View File
@@ -1,21 +0,0 @@
module.exports = {
extends: ["@commitlint/config-conventional"],
rules: {
// Configuration Format: [level, applicability, value]
// level: Error level, usually expressed as a number:
// 0 - disable rule
// 1 - Warning (does not prevent commits)
// 2 - Error (will block the commit)
// applicability: the conditions under which the rule applies, commonly used values:
// “always” - always apply the rule
// “never” - never apply the rule
// value: the specific value of the rule, e.g. a maximum length of 100.
// Refs: https://commitlint.js.org/reference/rules-configuration.html
"header-max-length": [2, "always", 100],
"type-enum": [
2,
"always",
["build", "chore", "ci", "docs", "feat", "fix", "perf", "refactor", "revert", "style", "test", "Release-As"]
]
}
};
.devcontainer/Dockerfile
-10
View File
@@ -1,10 +0,0 @@
# 1. Pull down your Azure Container Registry image
FROM rdagentappregistry.azurecr.io/rd-agent-mle:20250623
# 2. (Optional) install any additional tools you need
# e.g. git, bash-completion, etc.
# RUN apt update && \
# apt install -y git bash-completion && \
# rm -rf /var/lib/apt/lists/*
RUN apt update && \
apt install -y git bash-completion
.devcontainer/README.md
-39
View File
@@ -1,39 +0,0 @@
# Introduction
!!!!!This dev container is not for public development!!!!!!
!!!!!Please don't use it if you are just a public open-source user.!!!!!!
# Steps to run the dev container (for internal use only)
Prerequisites(this is the reason why this dev container is not for public use):
- Make sure you have the `rdagentappregistry.azurecr.io/rd-agent-mle:20250623` image locally & DevContainer is installed in your IDE
- The kaggle dataset is located at `/home/shared/RD-Agent/kaggle`
1. Open the project and select "Open In DevContainer"
2. Set up your Kaggle Key (do not share this; other internal URLs are hardcoded in the config files)
```bash
export KAGGLE_USERNAME=
export KAGGLE_KEY=
```
3. Run: python rdagent/app/data_science/loop.py --competition nomad2018-predict-transparent-conductors
# Additional Notes
- Please install and use this Dev Container in VS Code.
- You **must open VS Code remotely and enter the `RD-Agent` directory before running the DevContainer configuration (`.devcontainer/devcontainer.json`)**. Otherwise, the workspace and path mappings will not work as expected.
- To open the DevContainer correctly in VS Code:
1. Remotely connect to the machine and open the `RD-Agent` folder in VS Code.
2. Press `Ctrl+Shift+P` (or `Cmd+Shift+P` on Mac), type and select **"Dev Containers: Reopen in Container"**.
# How to grade your submission in the DevContainer
1. save your submission file in `./sumission.csv`
2. Run evaluation
DS_COMPETITION=<your competition name>
conda run -n mlebench mlebench grade-sample submission.csv $DS_COMPETITION --data-dir /tmp/kaggle/zip_files/
.devcontainer/env
-47
View File
@@ -1,47 +0,0 @@
# Global configs:
MAX_RETRY=12000
RETRY_WAIT_SECONDS=5
TIMEOUT_FAIL_LIMIT=100
# litellm
# CHAT_MODEL=gpt-4o
# CHAT_TEMPERATURE=0.7
CHAT_STREAM=False
CHAT_TEMPERATURE=1
CHAT_MODEL=o1-preview
SYSTEM_PROMPT_ROLE=user
BACKEND=rdagent.oai.backend.LiteLLMAPIBackend
OPENAI_API_KEY=sk-1234
OPENAI_API_BASE=http://ep14.213428.xyz:38881
# amc chat model configs:
EMBEDDING_MODEL=text-embedding-ada-002
# Cache Setting (Optional):
DUMP_CHAT_CACHE=True
USE_CHAT_CACHE=False
DUMP_EMBEDDING_CACHE=True
USE_EMBEDDING_CACHE=False
LOG_LLM_CHAT_CONTENT=True
DS_LOCAL_DATA_PATH=/tmp/kaggle
DS_IF_USING_MLE_DATA=True
PICKLE_CACHE_FOLDER_PATH_STR=./log/pickle_cache
CACHE_WITH_PICKLE=False
ENABLE_CACHE=False
PROMPT_CACHE_PATH=./log/prompt_cache.db
DS_CODER_COSTEER_ENV_TYPE=conda
# DS_PROPOSAL_VERSION=v2 deprecated
DS_CODER_ON_WHOLE_PIPELINE=True
COSTEER_V2_QUERY_FORMER_TRACE_LIMIT=3
# export PYTHONPATH=. # this is for running researcher branch;
.env.example
-61
View File
@@ -1,61 +0,0 @@
"""
This file is a template for the .env file.
Please copy this file to .env and fill in the values.
For more information about configuration options, please refer to the documentation
"""
# ==========================================
# Global configs:
MAX_RETRY=10
RETRY_WAIT_SECONDS=20
# ==========================================
# ==========================================
# Backend Configuration
# ==========================================
# BACKEND=rdagent.oai.backend.LiteLLMAPIBackend
# ==========================================
# ==========================================
# Backend Configuration (choose one)
# ==========================================
# 1. Set universal API key
# CHAT_MODEL="gpt-4o"
# EMBEDDING_MODEL="text-embedding-3-small"
# OPENAI_API_BASE="https://your-endpoint.com/v1"
# OPENAI_API_KEY="sk-your-api-key-here"
# 2. Set separate API KEY
# Chat configuration
OPENAI_API_KEY="sk-chat-key"
OPENAI_API_BASE="https://xxx-litellm.com/v1"
CHAT_MODEL='gpt-4o'
# Embedding configuration (using other service)
# Use siliconflow as example, pay attention to the litellm_proxy prefix
LITELLM_PROXY_API_KEY="sk-embedding-service-key"
LITELLM_PROXY_API_BASE="https://api.siliconflow.cn/v1"
EMBEDDING_MODEL="litellm_proxy/BAAI/bge-large-en-v1.5"
# ==========================================
# ==========================================
# Other Configuration
# ==========================================
# CHAT_AZURE_API_BASE=<for_Azure_user>
# CHAT_AZURE_API_VERSION=<for_Azure_user>
# EMBEDDING_AZURE_API_BASE=<for_Azure_user>
# EMBEDDING_AZURE_API_VERSION=<for_Azure_user>
# Cache Setting (Optional):
# USE_CHAT_CACHE=True
# USE_EMBEDDING_CACHE=True
# FT_DOCKER_ENABLE_CACHE=True
# DS_DOCKER_ENABLE_CACHE=True
# Senario Configs:
# ==========================================
.github/FUNDING.yml
-2
View File
@@ -1,2 +0,0 @@
github:
- MIIC-finance
.github/ISSUE_TEMPLATE/bug-report.md
-51
View File
@@ -1,51 +0,0 @@
---
name: "\U0001F41B Bug Report"
about: Submit a bug report to help us improve RD-Agent
labels: bug
---
## 🐛 Bug Description
<!-- A clear and concise description of what the bug is. -->
## To Reproduce
Steps to reproduce the behavior:
1.
2.
3.
## Expected Behavior
<!-- A clear and concise description of what you expected to happen. -->
## Screenshot
<!-- A screenshot of the error message or anything shouldn't appear-->
## Environment
**Note**: Users can run `rdagent collect_info` to get system information and paste it directly here.
- Name of current operating system:
- Processor architecture:
- System, version, and hardware information:
- Version number of the system:
- Python version:
- Container ID:
- Container Name:
- Container Status:
- Image ID used by the container:
- Image tag used by the container:
- Container port mapping:
- Container Label:
- Startup Commands:
- RD-Agent version:
- Package version:
## Additional Notes
<!-- Add any other information about the problem here. -->
.github/ISSUE_TEMPLATE/documentation.md
-9
View File
@@ -1,9 +0,0 @@
---
name: "\U0001F4D6 Documentation"
about: Report an issue related to documentation
---
## 📖 Documentation
<!-- Please specify whether it's tutorial part or API reference part, and describe it.-->
.github/ISSUE_TEMPLATE/feature-request.md
-25
View File
@@ -1,25 +0,0 @@
---
name: "\U0001F31FFeature Request"
about: Request for a new RD-Agent feature
labels: enhancement
---
## 🌟 Feature Description
<!-- A clear and concise description of the feature proposal -->
## Motivation
1. Application scenario
2. Related works (Papers, Github repos etc.):
3. Any other relevant and important information:
<!-- Please describe why the feature is important. -->
## Alternatives
<!-- A short description of any alternative solutions or features you've considered. -->
## Additional Notes
<!-- Add any other context or screenshots about the feature request here. -->
.github/ISSUE_TEMPLATE/question.md
-10
View File
@@ -1,10 +0,0 @@
---
name: "❓Questions & Help"
about: Have some questions? We can offer help.
labels: question
---
## ❓ Questions and Help
We sincerely suggest you to carefully read the [documentation](http://rdagent.readthedocs.io/). After that, if you still feel puzzled, please describe the question clearly under this issue.
.github/PULL_REQUEST_TEMPLATE.md
-34
View File
@@ -1,34 +0,0 @@
<!--- Thank you for submitting a Pull Request! In order to make our work smoother. -->
<!--- please make sure your Pull Request meets the following requirements: -->
<!--- 1. Provide a general summary of your changes in the Title above; -->
<!--- 2. Add appropriate prefixes to titles, such as `build:`, `chore:`, `ci:`, `docs:`, `feat:`, `fix:`, `perf:`, `refactor:`, `revert:`, `style:`, `test:`(Ref: https://www.conventionalcommits.org/). -->
<!--- Category: -->
<!--- Patch Updates: `fix:` -->
<!--- Example: fix(auth): correct login validation issue -->
<!--- minor update (introduces new functionality): `feat` -->
<!--- Example: feature(parser): add ability to parse arrays -->
<!--- major update(destructive update): Include BREAKING CHANGE in the commit message footer, or add `! ` in the commit footer to indicate that there is a destructive update. -->
<!--- Example: feat(auth)! : remove support for old authentication method -->
<!--- Other updates: `build:`, `chore:`, `ci:`, `docs:`, `perf:`, `refactor:`, `revert:`, `style:`, `test:`. -->
## Description
<!--- Describe your changes in detail -->
## Motivation and Context
<!--- Are there any related issues? If so, please put the link here. -->
<!--- Why is this change required? What problem does it solve? -->
## How Has This Been Tested?
<!--- Put an `x` in all the boxes that apply: --->
- [ ] If you are adding a new feature, test on your own test scripts.
<!--- **ATTENTION**: If you are adding a new feature, please make sure your codes are **correctly tested**. If our test scripts do not cover your cases, please provide your own test scripts under the `tests` folder and test them. More information about test scripts can be found [here](https://docs.python.org/3/library/unittest.html#basic-example), or you could refer to those we provide under the `tests` folder. -->
## Screenshots of Test Results (if appropriate):
1. Your own tests:
## Types of changes
<!--- What types of changes does your code introduce? Put an `x` in all the boxes that apply: -->
- [ ] Fix bugs
- [ ] Add new feature
- [ ] Update documentation
.github/dependabot.yml
-19
View File
@@ -1,19 +0,0 @@
updates:
- commit-message:
prefix: build(actions)
directory: /
package-ecosystem: github-actions
schedule:
interval: weekly
- commit-message:
prefix: build(requirements)
directory: /
groups:
dev:
dependency-type: development
prod:
dependency-type: production
package-ecosystem: pip
schedule:
interval: weekly
version: 2
.github/workflows/ci.yml
-69
View File
@@ -1,69 +0,0 @@
concurrency:
cancel-in-progress: true
group: ${{ github.workflow }}-${{ github.ref }}
jobs:
ci:
if: ${{ !cancelled() && ! failure() }}
needs: dependabot
runs-on: ubuntu-latest
steps:
- name: checkout
uses: actions/checkout@v4
with:
fetch-depth: 0
submodules: recursive
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v5
with:
cache: pip
python-version: ${{ matrix.python-version }}
- run: make dev
- name: lint test docs and build
run: make lint docs-gen test-offline # test docs build
strategy:
matrix:
python-version:
- '3.10'
- '3.11'
dependabot:
if: ${{ github.actor == 'dependabot[bot]' && startsWith(github.head_ref, 'dependabot/pip/') }}
permissions:
contents: write
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
with:
fetch-depth: 0
ref: ${{ github.head_ref }}
- name: Set up Git
run: |
git config --global user.name github-actions
git config --global user.email github-actions@github.com
- name: Set up Python with multiple versions.
uses: actions/setup-python@v5
with:
cache: pip
python-version: |
3.10
3.11
- name: Install pipenv using pipx
run: pipx install pipenv
- name: Generate constraints for all supported Python versions
run: |
CI= PYTHON_VERSION=3.10 make constraints
CI= PYTHON_VERSION=3.11 make constraints
- name: Push changes if applicable
run: |
if [[ -n `git status --porcelain` ]]; then
git commit -a -m "build: Update constraints for dependabot."
git push
fi
name: CI
on:
pull_request:
types:
- opened
- synchronize
push:
branches:
- main
.github/workflows/pr.yml
-35
View File
@@ -1,35 +0,0 @@
name: Lint pull request title
on:
pull_request:
types:
- opened
- synchronize
- reopened
- edited
concurrency:
cancel-in-progress: true
group: ${{ github.workflow }}-${{ github.ref }}
jobs:
lint-title:
runs-on: ubuntu-latest
steps:
# This step is necessary because the lint title uses the .commitlintrc.js file in the project root directory.
- name: Checkout Repository
uses: actions/checkout@v4
- name: Setup Node.js
uses: actions/setup-node@v4
with:
node-version: '16'
- name: Install commitlint
run: npm install --save-dev @commitlint/{config-conventional,cli}
- name: Validate PR Title with commitlint
env:
BODY: ${{ github.event.pull_request.title }}
run: |
echo "$BODY" | npx commitlint --config .commitlintrc.js
.github/workflows/readthedocs-preview.yml
-17
View File
@@ -1,17 +0,0 @@
concurrency:
cancel-in-progress: true
group: ${{ github.workflow }}-${{ github.ref }}
jobs:
documentation-links:
runs-on: ubuntu-latest
steps:
- uses: readthedocs/actions/preview@v1
with:
project-slug: RDAgent
name: Read the Docs Pull Request Preview
on:
pull_request_target:
types:
- opened
permissions:
pull-requests: write
.github/workflows/release.yml
-48
View File
@@ -1,48 +0,0 @@
name: Release
on:
push:
branches:
- main
permissions:
contents: read
jobs:
release_and_publish:
permissions:
contents: write
pull-requests: read
runs-on: ubuntu-latest
steps:
- name: Release please
id: release_please
uses: googleapis/release-please-action@v4
with:
# The current PAT (personal access token) was created on 2024-08-05,
# since the maximum validity of PAT is 1 year, you need to change the PAT before 2025-08-05.
token: ${{ secrets.PAT }}
release-type: simple
- uses: actions/checkout@v4
if: ${{ steps.release_please.outputs.release_created }}
with:
fetch-depth: 0
- name: Set up Python
if: ${{ steps.release_please.outputs.release_created }}
uses: actions/setup-python@v5
with:
cache: pip
python-version: '3.10'
- name: Install dependencies
if: ${{ steps.release_please.outputs.release_created }}
run: |
python -m pip install --upgrade pip
pip install setuptools wheel twine # better-exceptions(optional for debug)
- run: make dev
if: ${{ steps.release_please.outputs.release_created }}
- run: make build
if: ${{ steps.release_please.outputs.release_created }}
- name: upload
if: ${{ steps.release_please.outputs.release_created }}
env:
TWINE_USERNAME: __token__
TWINE_PASSWORD: ${{ secrets.PYPI_TOKEN }}
run: |
make upload
.gitignore
+11 -189
View File
@@ -1,190 +1,12 @@
# Custom
*.swp
.DS_Store
Pipfile
public
release-notes.md
typescript*
tmp/
.ai/
# Byte-compiled / optimized / DLL files
__pycache__/
*.py[cod]
*$py.class
# C extensions
*.so
# Distribution / packaging
.Python
build/
develop-eggs/
dist/
downloads/
eggs/
.eggs/
lib/
lib64/
parts/
sdist/
var/
wheels/
pip-wheel-metadata/
share/python-wheels/
*.egg-info/
.installed.cfg
*.egg
MANIFEST
# PyInstaller
# Usually these files are written by a python script from a template
# before PyInstaller builds the exe, so as to inject date/other infos into it.
*.manifest
*.spec
# Installer logs
pip-log.txt
pip-delete-this-directory.txt
# Unit test / coverage reports
htmlcov/
.tox/
.nox/
.coverage
.coverage.*
.cache
nosetests.xml
coverage.xml
*.cover
*.py,cover
.hypothesis/
.pytest_cache/
# Translations
*.mo
*.pot
# Django stuff:
*.log
/log*/
local_settings.py
db.sqlite3
db.sqlite3-journal
# Flask stuff:
instance/
.webassets-cache
# Scrapy stuff:
.scrapy
# Sphinx documentation
docs/_build/
# PyBuilder
target/
# Jupyter Notebook
.ipynb_checkpoints
# IPython
profile_default/
ipython_config.py
# pyenv
.python-version
# pipenv
# According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
# However, in case of collaboration, if having platform-specific dependencies or dependencies
# having no cross-platform support, pipenv may install dependencies that don't work, or not
# install all needed dependencies.
#Pipfile.lock
# PEP 582; used by e.g. github.com/David-OConnor/pyflow
__pypackages__/
# Celery stuff
celerybeat-schedule
celerybeat.pid
# SageMath parsed files
*.sage.py
# Environments
.env*
*.env
.venv
^env/
venv/
ENV/
env.bak/
venv.bak/
# Spyder project settings
.spyderproject
.spyproject
# Rope project settings
.ropeproject
# mkdocs documentation
/site
# mypy
.mypy_cache/
.dmypy.json
dmypy.json
# Pyre type checker
.pyre/
# all pkl files
*.pkl
# all h5 files
*.h5
# all vs-code files
.vscode/
# reports
reports/
# git_ignore_folder
.env
.env.backup*
*.backup*
git_ignore_folder/
#cache
*cache*/
*cache.json
# DB files
*.db
# Docker
factor_template/mlruns/
env_tpl
mlruns/
# possible output from coder or runner
*.pth
*qlib_res.csv
# shell script
*.out
/*.sh
.aider*
rdagent/app/benchmark/factor/example.json
# UI Server resources
videos/
static/
# AI assistant
.cursor/
.claude/
AGENTS.md
!rdagent/**/AGENTS.md
scripts/
fin_quant.log
selector.log
pickle_cache/
log/
__pycache__/
*.pyc
*.pyo
prompt_cache.db
.readthedocs.yaml
-38
View File
@@ -1,38 +0,0 @@
# .readthedocs.yml
# Read the Docs configuration file
# See https://docs.readthedocs.io/en/stable/config-file/v2.html for details
# Required
version: 2
# Set the version of Python and other tools you might need
build:
os: ubuntu-22.04
tools:
python: "3.10"
# During the build process, you need to fetch tags, and since the default command to read the docs only pulls shallow code, it will cause an error.
# So we added the `git fetch --tags --unshallow || true` command to fetch the full tag record.
# Adding this command overrides the default command, so we copied it over to make sure the build was successful.
commands:
- python -mvirtualenv $READTHEDOCS_VIRTUALENV_PATH
- python -m pip install --upgrade --no-cache-dir pip setuptools
- python -m pip install --upgrade --no-cache-dir sphinx
- python -m pip install --exists-action=w --no-cache-dir -r requirements/docs.txt
- python -m pip install --upgrade --upgrade-strategy only-if-needed --no-cache-dir .
- git fetch --tags --unshallow || true
- mkdir -p $READTHEDOCS_OUTPUT/html/
- python -m sphinx -T -b html -d _build/doctrees -D language=en ./docs $READTHEDOCS_OUTPUT/html
# Build documentation in the docs/ directory with Sphinx
sphinx:
configuration: docs/conf.py
# Build all formats
formats: all
# Optionally set the version of Python and requirements required to build your docs
python:
install:
- requirements: requirements/docs.txt
- method: pip
path: .
.streamlit/config.toml
-2
View File
@@ -1,2 +0,0 @@
[client]
showSidebarNavigation = false
LICENSE
+658 -17
View File
@@ -1,21 +1,662 @@
MIT License
GNU AFFERO GENERAL PUBLIC LICENSE
Version 3, 19 November 2007
Copyright (c) Microsoft Corporation.
Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
Everyone is permitted to copy and distribute verbatim copies
of this license document, but changing it is not allowed.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
Preamble
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
The GNU Affero General Public License is a free, copyleft license for
software and other kinds of works, specifically designed to ensure
cooperation with the community in the case of network server software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE
The licenses for most software and other practical works are designed
to take away your freedom to share and change the works. By contrast,
our General Public Licenses are intended to guarantee your freedom to
share and change all versions of a program--to make sure it remains free
software for all its users.
When we speak of free software, we are referring to freedom, not
price. Our General Public Licenses are designed to make sure that you
have the freedom to distribute copies of free software (and charge for
them if you wish), that you receive source code or can get it if you
want it, that you can change the software or use pieces of it in new
free programs, and that you know you can do these things.
Developers that use our General Public Licenses protect your rights
with two steps: (1) assert copyright on the software, and (2) offer
you this License which gives you legal permission to copy, distribute
and/or modify the software.
A secondary benefit of defending all users' freedom is that
improvements made in alternate versions of the program, if they
receive widespread use, become available for other developers to
incorporate. Many developers of free software are heartened and
encouraged by the resulting cooperation. However, in the case of
software used on network servers, this result may fail to come about.
The GNU General Public License permits making a modified version and
letting the public access it on a server without ever releasing its
source code to the public.
The GNU Affero General Public License is designed specifically to
ensure that, in such cases, the modified source code becomes available
to the community. It requires the operator of a network server to
provide the source code of the modified version running there to the
users of that server. Therefore, public use of a modified version, on
a publicly accessible server, gives the public access to the source
code of the modified version.
An older license, called the Affero General Public License and
published by Affero, was designed to accomplish similar goals. This is
a different license, not a version of the Affero GPL, but Affero has
released a new version of the Affero GPL which permits relicensing under
this license.
The precise terms and conditions for copying, distribution and
modification follow.
TERMS AND CONDITIONS
0. Definitions.
"This License" refers to version 3 of the GNU Affero General Public License.
"Copyright" also means copyright-like laws that apply to other kinds of
works, such as semiconductor masks.
"The Program" refers to any copyrightable work licensed under this
License. Each licensee is addressed as "you". "Licensees" and
"recipients" may be individuals or organizations.
To "modify" a work means to copy from or adapt all or part of the work
in a fashion requiring copyright permission, other than the making of an
exact copy. The resulting work is called a "modified version" of the
earlier work or a work "based on" the earlier work.
A "covered work" means either the unmodified Program or a work based
on the Program.
To "propagate" a work means to do anything with it that, without
permission, would make you directly or secondarily liable for
infringement under applicable copyright law, except executing it on a
computer or modifying a private copy. Propagation includes copying,
distribution (with or without modification), making available to the
public, and in some countries other activities as well.
To "convey" a work means any kind of propagation that enables other
parties to make or receive copies. Mere interaction with a user through
a computer network, with no transfer of a copy, is not conveying.
An interactive user interface displays "Appropriate Legal Notices"
to the extent that it includes a convenient and prominently visible
feature that (1) displays an appropriate copyright notice, and (2)
tells the user that there is no warranty for the work (except to the
extent that warranties are provided), that licensees may convey the
work under this License, and how to view a copy of this License. If
the interface presents a list of user commands or options, such as a
menu, a prominent item in the list meets this criterion.
1. Source Code.
The "source code" for a work means the preferred form of the work
for making modifications to it. "Object code" means any non-source
form of a work.
A "Standard Interface" means an interface that either is an official
standard defined by a recognized standards body, or, in the case of
interfaces specified for a particular programming language, one that
is widely used among developers working in that language.
The "System Libraries" of an executable work include anything, other
than the work as a whole, that (a) is included in the normal form of
packaging a Major Component, but which is not part of that Major
Component, and (b) serves only to enable use of the work with that
Major Component, or to implement a Standard Interface for which an
implementation is available to the public in source code form. A
"Major Component", in this context, means a major essential component
(kernel, window system, and so on) of the specific operating system
(if any) on which the executable work runs, or a compiler used to
produce the work, or an object code interpreter used to run it.
The "Corresponding Source" for a work in object code form means all
the source code needed to generate, install, and (for an executable
work) run the object code and to modify the work, including scripts to
control those activities. However, it does not include the work's
System Libraries, or general-purpose tools or generally available free
programs which are used unmodified in performing those activities but
which are not part of the work. For example, Corresponding Source
includes interface definition files associated with source files for
the work, and the source code for shared libraries and dynamically
linked subprograms that the work is specifically designed to require,
such as by intimate data communication or control flow between those
subprograms and other parts of the work.
The Corresponding Source need not include anything that users
can regenerate automatically from other parts of the Corresponding
Source.
The Corresponding Source for a work in source code form is that
same work.
2. Basic Permissions.
All rights granted under this License are granted for the term of
copyright on the Program, and are irrevocable provided the stated
conditions are met. This License explicitly affirms your unlimited
permission to run the unmodified Program. The output from running a
covered work is covered by this License only if the output, given its
content, constitutes a covered work. This License acknowledges your
rights of fair use or other equivalent, as provided by copyright law.
You may make, run and propagate covered works that you do not
convey, without conditions so long as your license otherwise remains
in force. You may convey covered works to others for the sole purpose
of having them make modifications exclusively for you, or provide you
with facilities for running those works, provided that you comply with
the terms of this License in conveying all material for which you do
not control copyright. Those thus making or running the covered works
for you must do so exclusively on your behalf, under your direction
and control, on terms that prohibit them from making any copies of
your copyrighted material outside their relationship with you.
Conveying under any other circumstances is permitted solely under
the conditions stated below. Sublicensing is not allowed; section 10
makes it unnecessary.
3. Protecting Users' Legal Rights From Anti-Circumvention Law.
No covered work shall be deemed part of an effective technological
measure under any applicable law fulfilling obligations under article
11 of the WIPO copyright treaty adopted on 20 December 1996, or
similar laws prohibiting or restricting circumvention of such
measures.
When you convey a covered work, you waive any legal power to forbid
circumvention of technological measures to the extent such circumvention
is effected by exercising rights under this License with respect to
the covered work, and you disclaim any intention to limit operation or
modification of the work as a means of enforcing, against the work's
users, your or third parties' legal rights to forbid circumvention of
technological measures.
4. Conveying Verbatim Copies.
You may convey verbatim copies of the Program's source code as you
receive it, in any medium, provided that you conspicuously and
appropriately publish on each copy an appropriate copyright notice;
keep intact all notices stating that this License and any
non-permissive terms added in accord with section 7 apply to the code;
keep intact all notices of the absence of any warranty; and give all
recipients a copy of this License along with the Program.
You may charge any price or no price for each copy that you convey,
and you may offer support or warranty protection for a fee.
5. Conveying Modified Source Versions.
You may convey a work based on the Program, or the modifications to
produce it from the Program, in the form of source code under the
terms of section 4, provided that you also meet all of these conditions:
a) The work must carry prominent notices stating that you modified
it, and giving a relevant date.
b) The work must carry prominent notices stating that it is
released under this License and any conditions added under section
7. This requirement modifies the requirement in section 4 to
"keep intact all notices".
c) You must license the entire work, as a whole, under this
License to anyone who comes into possession of a copy. This
License will therefore apply, along with any applicable section 7
additional terms, to the whole of the work, and all its parts,
regardless of how they are packaged. This License gives no
permission to license the work in any other way, but it does not
invalidate such permission if you have separately received it.
d) If the work has interactive user interfaces, each must display
Appropriate Legal Notices; however, if the Program has interactive
interfaces that do not display Appropriate Legal Notices, your
work need not make them do so.
A compilation of a covered work with other separate and independent
works, which are not by their nature extensions of the covered work,
and which are not combined with it such as to form a larger program,
in or on a volume of a storage or distribution medium, is called an
"aggregate" if the compilation and its resulting copyright are not
used to limit the access or legal rights of the compilation's users
beyond what the individual works permit. Inclusion of a covered work
in an aggregate does not cause this License to apply to the other
parts of the aggregate.
6. Conveying Non-Source Forms.
You may convey a covered work in object code form under the terms
of sections 4 and 5, provided that you also convey the
machine-readable Corresponding Source under the terms of this License,
in one of these ways:
a) Convey the object code in, or embodied in, a physical product
(including a physical distribution medium), accompanied by the
Corresponding Source fixed on a durable physical medium
customarily used for software interchange.
b) Convey the object code in, or embodied in, a physical product
(including a physical distribution medium), accompanied by a
written offer, valid for at least three years and valid for as
long as you offer spare parts or customer support for that product
model, to give anyone who possesses the object code either (1) a
copy of the Corresponding Source for all the software in the
product that is covered by this License, on a durable physical
medium customarily used for software interchange, for a price no
more than your reasonable cost of physically performing this
conveying of source, or (2) access to copy the
Corresponding Source from a network server at no charge.
c) Convey individual copies of the object code with a copy of the
written offer to provide the Corresponding Source. This
alternative is allowed only occasionally and noncommercially, and
only if you received the object code with such an offer, in accord
with subsection 6b.
d) Convey the object code by offering access from a designated
place (gratis or for a charge), and offer equivalent access to the
Corresponding Source in the same way through the same place at no
further charge. You need not require recipients to copy the
Corresponding Source along with the object code. If the place to
copy the object code is a network server, the Corresponding Source
may be on a different server (operated by you or a third party)
that supports equivalent copying facilities, provided you maintain
clear directions next to the object code saying where to find the
Corresponding Source. Regardless of what server hosts the
Corresponding Source, you remain obligated to ensure that it is
available for as long as needed to satisfy these requirements.
e) Convey the object code using peer-to-peer transmission, provided
you inform other peers where the object code and Corresponding
Source of the work are being offered to the general public at no
charge under subsection 6d.
A separable portion of the object code, whose source code is excluded
from the Corresponding Source as a System Library, need not be
included in conveying the object code work.
A "User Product" is either (1) a "consumer product", which means any
tangible personal property which is normally used for personal, family,
or household purposes, or (2) anything designed or sold for incorporation
into a dwelling. In determining whether a product is a consumer product,
doubtful cases shall be resolved in favor of coverage. For a particular
product received by a particular user, "normally used" refers to a
typical or common use of that class of product, regardless of the status
of the particular user or of the way in which the particular user
actually uses, or expects or is expected to use, the product. A product
is a consumer product regardless of whether the product has substantial
commercial, industrial or non-consumer uses, unless such uses represent
the only significant mode of use of the product.
"Installation Information" for a User Product means any methods,
procedures, authorization keys, or other information required to install
and execute modified versions of a covered work in that User Product from
a modified version of its Corresponding Source. The information must
suffice to ensure that the continued functioning of the modified object
code is in no case prevented or interfered with solely because
modification has been made.
If you convey an object code work under this section in, or with, or
specifically for use in, a User Product, and the conveying occurs as
part of a transaction in which the right of possession and use of the
User Product is transferred to the recipient in perpetuity or for a
fixed term (regardless of how the transaction is characterized), the
Corresponding Source conveyed under this section must be accompanied
by the Installation Information. But this requirement does not apply
if neither you nor any third party retains the ability to install
modified object code on the User Product (for example, the work has
been installed in ROM).
The requirement to provide Installation Information does not include a
requirement to continue to provide support service, warranty, or updates
for a work that has been modified or installed by the recipient, or for
the User Product in which it has been modified or installed. Access to a
network may be denied when the modification itself materially and
adversely affects the operation of the network or violates the rules and
protocols for communication across the network.
Corresponding Source conveyed, and Installation Information provided,
in accord with this section must be in a format that is publicly
documented (and with an implementation available to the public in
source code form), and must require no special password or key for
unpacking, reading or copying.
7. Additional Terms.
"Additional permissions" are terms that supplement the terms of this
License by making exceptions from one or more of its conditions.
Additional permissions that are applicable to the entire Program shall
be treated as though they were included in this License, to the extent
that they are valid under applicable law. If additional permissions
apply only to part of the Program, that part may be used separately
under those permissions, but the entire Program remains governed by
this License without regard to the additional permissions.
When you convey a copy of a covered work, you may at your option
remove any additional permissions from that copy, or from any part of
it. (Additional permissions may be written to require their own
removal in certain cases when you modify the work.) You may place
additional permissions on material, added by you to a covered work,
for which you have or can give appropriate copyright permission.
Notwithstanding any other provision of this License, for material you
add to a covered work, you may (if authorized by the copyright holders of
that material) supplement the terms of this License with terms:
a) Disclaiming warranty or limiting liability differently from the
terms of sections 15 and 16 of this License; or
b) Requiring preservation of specified reasonable legal notices or
author attributions in that material or in the Appropriate Legal
Notices displayed by works containing it; or
c) Prohibiting misrepresentation of the origin of that material, or
requiring that modified versions of such material be marked in
reasonable ways as different from the original version; or
d) Limiting the use for publicity purposes of names of licensors or
authors of the material; or
e) Declining to grant rights under trademark law for use of some
trade names, trademarks, or service marks; or
f) Requiring indemnification of licensors and authors of that
material by anyone who conveys the material (or modified versions of
it) with contractual assumptions of liability to the recipient, for
any liability that these contractual assumptions directly impose on
those licensors and authors.
All other non-permissive additional terms are considered "further
restrictions" within the meaning of section 10. If the Program as you
received it, or any part of it, contains a notice stating that it is
governed by this License along with a term that is a further
restriction, you may remove that term. If a license document contains
a further restriction but permits relicensing or conveying under this
License, you may add to a covered work material governed by the terms
of that license document, provided that the further restriction does
not survive such relicensing or conveying.
If you add terms to a covered work in accord with this section, you
must place, in the relevant source files, a statement of the
additional terms that apply to those files, or a notice indicating
where to find the applicable terms.
Additional terms, permissive or non-permissive, may be stated in the
form of a separately written license, or stated as exceptions;
the above requirements apply either way.
8. Termination.
You may not propagate or modify a covered work except as expressly
provided under this License. Any attempt otherwise to propagate or
modify it is void, and will automatically terminate your rights under
this License (including any patent licenses granted under the third
paragraph of section 11).
However, if you cease all violation of this License, then your
license from a particular copyright holder is reinstated (a)
provisionally, unless and until the copyright holder explicitly and
finally terminates your license, and (b) permanently, if the copyright
holder fails to notify you of the violation by some reasonable means
prior to 60 days after the cessation.
Moreover, your license from a particular copyright holder is
reinstated permanently if the copyright holder notifies you of the
violation by some reasonable means, this is the first time you have
received notice of violation of this License (for any work) from that
copyright holder, and you cure the violation prior to 30 days after
your receipt of the notice.
Termination of your rights under this section does not terminate the
licenses of parties who have received copies or rights from you under
this License. If your rights have been terminated and not permanently
reinstated, you do not qualify to receive new licenses for the same
material under section 10.
9. Acceptance Not Required for Having Copies.
You are not required to accept this License in order to receive or
run a copy of the Program. Ancillary propagation of a covered work
occurring solely as a consequence of using peer-to-peer transmission
to receive a copy likewise does not require acceptance. However,
nothing other than this License grants you permission to propagate or
modify any covered work. These actions infringe copyright if you do
not accept this License. Therefore, by modifying or propagating a
covered work, you indicate your acceptance of this License to do so.
10. Automatic Licensing of Downstream Recipients.
Each time you convey a covered work, the recipient automatically
receives a license from the original licensors, to run, modify and
propagate that work, subject to this License. You are not responsible
for enforcing compliance by third parties with this License.
An "entity transaction" is a transaction transferring control of an
organization, or substantially all assets of one, or subdividing an
organization, or merging organizations. If propagation of a covered
work results from an entity transaction, each party to that
transaction who receives a copy of the work also receives whatever
licenses to the work the party's predecessor in interest had or could
give under the previous paragraph, plus a right to possession of the
Corresponding Source of the work from the predecessor in interest, if
the predecessor has it or can get it with reasonable efforts.
You may not impose any further restrictions on the exercise of the
rights granted or affirmed under this License. For example, you may
not impose a license fee, royalty, or other charge for exercise of
rights granted under this License, and you may not initiate litigation
(including a cross-claim or counterclaim in a lawsuit) alleging that
any patent claim is infringed by making, using, selling, offering for
sale, or importing the Program or any portion of it.
11. Patents.
A "contributor" is a copyright holder who authorizes use under this
License of the Program or a work on which the Program is based. The
work thus licensed is called the contributor's "contributor version".
A contributor's "essential patent claims" are all patent claims
owned or controlled by the contributor, whether already acquired or
hereafter acquired, that would be infringed by some manner, permitted
by this License, of making, using, or selling its contributor version,
but do not include claims that would be infringed only as a
consequence of further modification of the contributor version. For
purposes of this definition, "control" includes the right to grant
patent sublicenses in a manner consistent with the requirements of
this License.
Each contributor grants you a non-exclusive, worldwide, royalty-free
patent license under the contributor's essential patent claims, to
make, use, sell, offer for sale, import and otherwise run, modify and
propagate the contents of its contributor version.
In the following three paragraphs, a "patent license" is any express
agreement or commitment, however denominated, not to enforce a patent
(such as an express permission to practice a patent or covenant not to
sue for patent infringement). To "grant" such a patent license to a
party means to make such an agreement or commitment not to enforce a
patent against the party.
If you convey a covered work, knowingly relying on a patent license,
and the Corresponding Source of the work is not available for anyone
to copy, free of charge and under the terms of this License, through a
publicly available network server or other readily accessible means,
then you must either (1) cause the Corresponding Source to be so
available, or (2) arrange to deprive yourself of the benefit of the
patent license for this particular work, or (3) arrange, in a manner
consistent with the requirements of this License, to extend the patent
license to downstream recipients. "Knowingly relying" means you have
actual knowledge that, but for the patent license, your conveying the
covered work in a country, or your recipient's use of the covered work
in a country, would infringe one or more identifiable patents in that
country that you have reason to believe are valid.
If, pursuant to or in connection with a single transaction or
arrangement, you convey, or propagate by procuring conveyance of, a
covered work, and grant a patent license to some of the parties
receiving the covered work authorizing them to use, propagate, modify
or convey a specific copy of the covered work, then the patent license
you grant is automatically extended to all recipients of the covered
work and works based on it.
A patent license is "discriminatory" if it does not include within
the scope of its coverage, prohibits the exercise of, or is
conditioned on the non-exercise of one or more of the rights that are
specifically granted under this License. You may not convey a covered
work if you are a party to an arrangement with a third party that is
in the business of distributing software, under which you make payment
to the third party based on the extent of your activity of conveying
the work, and under which the third party grants, to any of the
parties who would receive the covered work from you, a discriminatory
patent license (a) in connection with copies of the covered work
conveyed by you (or copies made from those copies), or (b) primarily
for and in connection with specific products or compilations that
contain the covered work, unless you entered into that arrangement,
or that patent license was granted, prior to 28 March 2007.
Nothing in this License shall be construed as excluding or limiting
any implied license or other defenses to infringement that may
otherwise be available to you under applicable patent law.
12. No Surrender of Others' Freedom.
If conditions are imposed on you (whether by court order, agreement or
otherwise) that contradict the conditions of this License, they do not
excuse you from the conditions of this License. If you cannot convey a
covered work so as to satisfy simultaneously your obligations under this
License and any other pertinent obligations, then as a consequence you may
not convey it at all. For example, if you agree to terms that obligate you
to collect a royalty for further conveying from those to whom you convey
the Program, the only way you could satisfy both those terms and this
License would be to refrain entirely from conveying the Program.
13. Remote Network Interaction; Use with the GNU General Public License.
Notwithstanding any other provision of this License, if you modify the
Program, your modified version must prominently offer all users
interacting with it remotely through a computer network (if your version
supports such interaction) an opportunity to receive the Corresponding
Source of your version by providing access to the Corresponding Source
from a network server at no charge, through some standard or customary
means of facilitating copying of software. This Corresponding Source
shall include the Corresponding Source for any work covered by version 3
of the GNU General Public License that is incorporated pursuant to the
following paragraph.
Notwithstanding any other provision of this License, you have
permission to link or combine any covered work with a work licensed
under version 3 of the GNU General Public License into a single
combined work, and to convey the resulting work. The terms of this
License will continue to apply to the part which is the covered work,
but the work with which it is combined will remain governed by version
3 of the GNU General Public License.
14. Revised Versions of this License.
The Free Software Foundation may publish revised and/or new versions of
the GNU Affero General Public License from time to time. Such new versions
will be similar in spirit to the present version, but may differ in detail to
address new problems or concerns.
Each version is given a distinguishing version number. If the
Program specifies that a certain numbered version of the GNU Affero General
Public License "or any later version" applies to it, you have the
option of following the terms and conditions either of that numbered
version or of any later version published by the Free Software
Foundation. If the Program does not specify a version number of the
GNU Affero General Public License, you may choose any version ever published
by the Free Software Foundation.
If the Program specifies that a proxy can decide which future
versions of the GNU Affero General Public License can be used, that proxy's
public statement of acceptance of a version permanently authorizes you
to choose that version for the Program.
Later license versions may give you additional or different
permissions. However, no additional obligations are imposed on any
author or copyright holder as a result of your choosing to follow a
later version.
15. Disclaimer of Warranty.
THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
16. Limitation of Liability.
IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
SUCH DAMAGES.
17. Interpretation of Sections 15 and 16.
If the disclaimer of warranty and limitation of liability provided
above cannot be given local legal effect according to their terms,
reviewing courts shall apply local law that most closely approximates
an absolute waiver of all civil liability in connection with the
Program, unless a warranty or assumption of liability accompanies a
copy of the Program in return for a fee.
END OF TERMS AND CONDITIONS
How to Apply These Terms to Your New Programs
If you develop a new program, and you want it to be of the greatest
possible use to the public, the best way to achieve this is to make it
free software which everyone can redistribute and change under these terms.
To do so, attach the following notices to the program. It is safest
to attach them to the start of each source file to most effectively
state the exclusion of warranty; and each file should have at least
the "copyright" line and a pointer to where the full notice is found.
<one line to give the program's name and a brief idea of what it does.>
Copyright (C) {{ year }} {{ organization }}
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Affero General Public License for more details.
You should have received a copy of the GNU Affero General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
Also add information on how to contact you by electronic and paper mail.
If your software can interact with users remotely through a computer
network, you should also make sure that it provides a way for users to
get its source. For example, if your program is a web application, its
interface could display a "Source" link that leads users to an archive
of the code. There are many ways you could offer source, and different
solutions will be better for different programs; see section 13 for the
specific requirements.
You should also get your employer (if you work as a programmer) or school,
if any, to sign a "copyright disclaimer" for the program, if necessary.
For more information on this, and how to apply and follow the GNU AGPL, see
<http://www.gnu.org/licenses/>.
rdagent/app/qlib_rd_loop/prompts.yaml
+46 -6
View File
@@ -1,15 +1,55 @@
hypothesis_generation:
system: |-
You are an expert in financial analysis. Your task is to generate a well-reasoned hypothesis based on the provided financial factors and report content.
Please ensure your response is in JSON format as shown below:
You are an expert quantitative researcher specialized in FX (foreign exchange) trading,
specifically EURUSD intraday strategies on 15-minute bars.
EURUSD domain knowledge you must apply:
- London session (08:00-12:00 UTC): highest volatility, trending behavior — favor momentum strategies
- NY session (13:00-17:00 UTC): second volatility peak, also trending
- Asian session (00:00-07:00 UTC): low volatility, mean-reverting behavior
- London/NY overlap (13:00-17:00 UTC): strongest directional moves of the day
- Weekend gap risk: avoid holding positions after Friday 20:00 UTC
- Spread cost: ~1.5 bps per trade — strategies must minimize unnecessary entries
- EURUSD is mean-reverting on short windows (<1h), trending on longer (>4h)
- Key macro drivers: ECB/Fed rate decisions, NFP (first Friday of month), CPI releases
Available model types you can propose:
- TimeSeries: LSTM, GRU, TCN (Temporal Convolutional Network), Transformer, PatchTST
- Tabular: XGBoost, LightGBM, RandomForest (on engineered features)
- Hybrid: CNN+LSTM, XGBoost+LSTM ensemble
- Statistical: Regime-switching (HMM), Kalman filter
Available features in the dataset:
- OHLCV: open, high, low, close, volume (15min bars)
- Returns: ret_1, ret_4, ret_8, ret_16, ret_96
- Technical: rsi_14, macd_hist, adx_14, atr_14, bb_pct, stoch_k, cci_14
- Volatility: vol_real_4, vol_real_16, vol_ratio, zscore_ret_96
- Time/Session: hour, is_london, is_ny, is_overlap, hour_sin, hour_cos
- Lags: rsi_14_lag1-8, macd_hist_lag1-8, bb_pct_lag1-8
Your hypothesis must:
1. Specify which session(s) the strategy targets
2. Name which model type to use and why it fits EURUSD
3. Include a session filter (is_london / is_ny)
4. Include a spread filter (only trade when expected |return| > 0.0003)
5. Specify target: classification (fwd_sign_4) or regression (fwd_ret_4)
Please ensure your response is in JSON format:
{
"hypothesis": "A clear and concise hypothesis based on the provided information.",
"reason": "A detailed explanation supporting the generated hypothesis.",
"hypothesis": "A clear and concise trading hypothesis for EURUSD 15min.",
"reason": "Detailed explanation including session, model choice, and expected edge.",
"model_type": "One of: TimeSeries / Tabular / XGBoost",
"target_session": "london / ny / asian / all",
"expected_arr_range": "e.g. 8-12%"
}
user: |-
The following are the financial factors and their descriptions:
Previously tried approaches and their results:
{{ factor_descriptions }}
The report content is as follows:
Additional context:
{{ report_content }}
Generate a NEW hypothesis that is meaningfully different from what has been tried.
Focus on approaches that have NOT been tested yet.
Target: beat current best ARR of 9.62%.
rdagent/components/coder/model_coder/prompts.yaml
+9
View File
@@ -31,6 +31,15 @@ evolving_strategy_model_coder:
system: |-
User is trying to implement some pytorch models in the following scenario:
{{ scenario }}
EURUSD-specific rules (ALWAYS apply these in generated code):
1. Session filter: use is_london and is_ny columns — weight/filter signals to active sessions
2. Spread filter: only generate signal when abs(predicted_return) > 0.0003
3. ADX regime: if adx_proxy > 1.2 use trend model; if adx_proxy < 0.8 use mean-reversion
4. Weekend filter: zero out signals when dayofweek==4 and hour>=20
5. Max trade frequency: target <15 trades per day (avoid spread cost death)
6. Supported model_type values: "Tabular", "TimeSeries", "XGBoost"
Your code is expected to align the scenario in any form which means The user needs to get the prediction of the model based on the input data.
To help you write the correct code, the user might provide multiple information that helps you write the correct code:
rdagent/log/__init__.py
-4
View File
@@ -1,4 +0,0 @@
from rdagent.log.logger import RDAgentLog
from rdagent.log.utils import LogColors
rdagent_logger: RDAgentLog = RDAgentLog()
rdagent/log/base.py
-103
View File
@@ -1,103 +0,0 @@
from __future__ import annotations
from abc import abstractmethod
from collections.abc import Generator
from dataclasses import dataclass
from datetime import datetime
from pathlib import Path
from typing import Literal, Optional
@dataclass
class Message:
"""The info unit of the storage"""
tag: str # namespace like like a.b.c
level: Literal["DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"] # The level of the logging
timestamp: datetime # The time when the message is generated
caller: Optional[
str
] # The caller of the logging like `rdagent.oai.llm_utils:_create_chat_completion_inner_function:55`(file:func:line)
pid_trace: Optional[str] # The process id trace; A-B-C represents A create B, B create C
content: object # The content
class Storage:
"""
Basic storage to support saving objects;
# Usage:
The storage has mainly two kind of users:
- The logging end: you can choose any of the following method to use the object
- We can use it directly with the native logging storage
- We can use it with other logging tools; For example, serve as a handler for loggers
- The view end:
- Mainly for the subclass of `logging.base.View`
- It should provide two kind of ways to provide content
- offline content provision.
- online content preovision.
"""
@abstractmethod
def log(
self,
obj: object,
tag: str = "",
timestamp: datetime | None = None,
) -> str | Path:
"""
Parameters
----------
obj : object
The object for logging.
name : str
The name of the object. For example "a.b.c"
We may log a lot of objects to a same name
Returns
-------
str | Path
The storage identifier of the object.
"""
...
@abstractmethod
def iter_msg(self) -> Generator[Message, None, None]:
"""
Iterate the message in the storage.
"""
...
@abstractmethod
def truncate(self, time: datetime) -> None:
"""
Remove all log entries after the specified time.
"""
...
def __str__(self) -> str:
return self.__class__.__name__
class View:
"""
Motivation:
Display the content in the storage
"""
# TODO: pleas fix me
@abstractmethod
def display(self, s: Storage, watch: bool = False) -> None:
"""
Parameters
----------
s : Storage
watch : bool
should we watch the new content and display them
"""
...
rdagent/log/conf.py
-34
View File
@@ -1,34 +0,0 @@
from datetime import datetime, timezone
from pathlib import Path
from typing import Any
from pydantic_settings import SettingsConfigDict
from rdagent.core.conf import ExtendedBaseSettings
class LogSettings(ExtendedBaseSettings):
model_config = SettingsConfigDict(env_prefix="LOG_", protected_namespaces=())
trace_path: str = str(Path.cwd() / "log" / datetime.now(timezone.utc).strftime("%Y-%m-%d_%H-%M-%S-%f"))
format_console: str | None = None
""""If it is None, leave it as the default"""
ui_server_port: int | None = None
storages: dict[str, list[int | str]] = {}
def set_ui_server_port(self, port: int | None) -> None:
self.ui_server_port = port
if port is None:
self.storages.pop("rdagent.log.ui.storage.WebStorage", None)
return
self.storages["rdagent.log.ui.storage.WebStorage"] = [port, self.trace_path]
def model_post_init(self, _context: Any, /) -> None:
self.set_ui_server_port(self.ui_server_port)
LOG_SETTINGS = LogSettings()
rdagent/log/logger.py
-153
View File
@@ -1,153 +0,0 @@
import os
import sys
from contextlib import contextmanager
from contextvars import ContextVar
from datetime import datetime
from pathlib import Path
from typing import Generator
from loguru import logger
from psutil import Process
from rdagent.core.utils import SingletonBaseClass, import_class
from .base import Storage
from .conf import LOG_SETTINGS
from .storage import FileStorage
from .utils import get_caller_info
class RDAgentLog(SingletonBaseClass):
"""
The files are organized based on the tag & PID
Here is an example tag
.. code-block::
a
- b
- c
- 123
- common_logs.log
- 1322
- common_logs.log
- 1233
- <timestamp>.pkl
- d
- 1233-673 ...
- 1233-4563 ...
- 1233-365 ...
"""
# Thread-/coroutine-local tag; In Linux forked subprocess, it will be copied to the subprocess.
_tag_ctx: ContextVar[str] = ContextVar("_tag_ctx", default="")
_raw_log_key = "_rdagent_raw"
@classmethod
def _configure_console_sinks(cls) -> None:
raw_filter = lambda record: bool(record["extra"].get(cls._raw_log_key, False))
normal_filter = lambda record: not raw_filter(record)
if LOG_SETTINGS.format_console is not None:
logger.add(sys.stdout, format=LOG_SETTINGS.format_console, filter=normal_filter)
else:
logger.add(sys.stdout, filter=normal_filter)
logger.add(sys.stdout, format="{message}", filter=raw_filter)
@property
def _tag(self) -> str: # Get current tag
return self._tag_ctx.get()
@_tag.setter # Set current tag
def _tag(self, value: str) -> None:
self._tag_ctx.set(value)
def __init__(self) -> None:
logger.remove()
self._configure_console_sinks()
self.storage = FileStorage(LOG_SETTINGS.trace_path)
self.other_storages: list[Storage] = []
self.refresh_storages_from_settings()
self.main_pid = os.getpid()
def refresh_storages_from_settings(self) -> None:
self.other_storages = []
for storage, args in LOG_SETTINGS.storages.items():
storage_cls = import_class(storage)
self.other_storages.append(storage_cls(*args))
def rebind_console_to_current_streams(self) -> None:
"""Rebind loguru sinks to the current stdio objects.
This is needed in forked/spawned subprocesses after stdout/stderr have been
redirected, because loguru keeps references to the original stream objects.
"""
logger.remove()
self._configure_console_sinks()
@contextmanager
def tag(self, tag: str) -> Generator[None, None, None]:
if tag.strip() == "":
raise ValueError("Tag cannot be empty.")
# Generate a new complete tag
current_tag = self._tag_ctx.get()
new_tag = tag if current_tag == "" else f"{current_tag}.{tag}"
# Set and save token for later restore
token = self._tag_ctx.set(new_tag)
try:
yield
finally:
# Restore previous tag (thread/coroutine safe)
self._tag_ctx.reset(token)
def set_storages_path(self, path: str | Path) -> None:
if isinstance(path, str):
path = Path(path)
for storage in [self.storage] + self.other_storages:
if hasattr(storage, "path"):
storage.path = path
def truncate_storages(self, time: datetime) -> None:
for storage in [self.storage] + self.other_storages:
storage.truncate(time=time)
def get_pids(self) -> str:
"""
Returns a string of pids from the current process to the main process.
Split by '-'.
"""
pid = os.getpid()
process = Process(pid)
pid_chain = f"{pid}"
while process.pid != self.main_pid:
parent_pid = process.ppid()
parent_process = Process(parent_pid)
pid_chain = f"{parent_pid}-{pid_chain}"
process = parent_process
return pid_chain
def log_object(self, obj: object, *, tag: str = "") -> None:
tag = f"{self._tag}.{tag}.{self.get_pids()}".strip(".")
for storage in [self.storage] + self.other_storages:
storage.log(obj, tag=tag)
def _log(self, level: str, msg: str, *, tag: str = "", raw: bool = False) -> None:
caller_info = get_caller_info(level=3)
tag = f"{self._tag}.{tag}.{self.get_pids()}".strip(".")
patched_logger = logger.patch(lambda r: r.update(caller_info)).bind(**{self._raw_log_key: raw}).opt(raw=raw)
log_func = getattr(patched_logger, level)
log_func(msg)
def info(self, msg: str, *, tag: str = "", raw: bool = False) -> None:
self._log("info", msg, tag=tag, raw=raw)
def warning(self, msg: str, *, tag: str = "", raw: bool = False) -> None:
self._log("warning", msg, tag=tag, raw=raw)
def error(self, msg: str, *, tag: str = "", raw: bool = False) -> None:
self._log("error", msg, tag=tag, raw=raw)
rdagent/log/mle_summary.py
-260
View File
@@ -1,260 +0,0 @@
import pickle
import traceback
from collections import defaultdict
from pathlib import Path
import fire
import pandas as pd
from rdagent.core.experiment import FBWorkspace
from rdagent.core.proposal import ExperimentFeedback
from rdagent.log.storage import FileStorage
from rdagent.log.utils import extract_json, extract_loopid_func_name, is_valid_session
from rdagent.log.utils.folder import get_first_session_file_after_duration
from rdagent.scenarios.data_science.experiment.experiment import DSExperiment
from rdagent.scenarios.data_science.test_eval import (
MLETestEval,
NoTestEvalError,
get_test_eval,
)
# from rdagent.scenarios.kaggle.kaggle_crawler import score_rank
from rdagent.utils.workflow import LoopBase
def save_grade_info(log_trace_path: Path):
test_eval = get_test_eval()
trace_storage = FileStorage(log_trace_path)
for msg in trace_storage.iter_msg(tag="competition"):
competition = msg.content
for msg in trace_storage.iter_msg(tag="running"):
if isinstance(msg.content, DSExperiment):
# TODO: mle_score.txt is not a general name now.
# Please use a more general name like test_score.txt
try:
mle_score_str = test_eval.eval(competition, msg.content.experiment_workspace)
trace_storage.log(
mle_score_str, tag=f"{msg.tag}.mle_score.pid", save_type="pkl", timestamp=msg.timestamp
)
except Exception as e:
print(f"Error in {log_trace_path}: {e}", traceback.format_exc())
def save_all_grade_info(log_folder: str | Path) -> None:
for log_trace_path in Path(log_folder).iterdir():
if is_valid_session(log_trace_path):
try:
save_grade_info(log_trace_path)
except NoTestEvalError as e:
print(f"Error in {log_trace_path}: {e}", traceback.format_exc())
def _get_loop_and_fn_after_hours(log_folder: Path, hours: int):
stop_session_fp = get_first_session_file_after_duration(log_folder, f"{hours}h")
with stop_session_fp.open("rb") as f:
session_obj: LoopBase = pickle.load(f)
loop_trace = session_obj.loop_trace
stop_li = max(loop_trace.keys())
last_loop = loop_trace[stop_li]
last_step = last_loop[-1]
stop_fn = session_obj.steps[last_step.step_idx]
print(f"Stop Loop: {stop_li=}, {stop_fn=}")
files = sorted(
(log_folder / "__session__").glob("*/*_*"), key=lambda f: (int(f.parent.name), int(f.name.split("_")[0]))
)
print(f"Max Session: {files[-1:]=}")
return stop_li, stop_fn
def summarize_folder(log_folder: Path, hours: int | None = None) -> None:
test_eval = get_test_eval()
is_mle = isinstance(test_eval, MLETestEval)
"""
Summarize the log folder and save the summary as a pickle file.
Args:
log_folder (Path): The path to the log folder (contains many log traces).
hours (int | None): The number of hours to stat. If None, stat all.
"""
log_folder = Path(log_folder)
stat = defaultdict(dict)
for log_trace_path in log_folder.iterdir(): # One log trace
if not is_valid_session(log_trace_path):
continue
loop_num = 0
made_submission_num = 0
valid_submission_num = 0
above_median_num = 0
get_medal_num = 0
bronze_num = 0
silver_num = 0
gold_num = 0
test_scores = {}
test_ranks = {}
valid_scores = {}
bronze_threshold = 0.0
silver_threshold = 0.0
gold_threshold = 0.0
median_threshold = 0.0
success_loop_num = 0
sota_exp_stat = ""
sota_exp_score = None
sota_exp_rank = None
grade_output = None
if hours:
stop_li, stop_fn = _get_loop_and_fn_after_hours(log_trace_path, hours)
msgs = [(msg, extract_loopid_func_name(msg.tag)) for msg in FileStorage(log_trace_path).iter_msg()]
msgs = [(msg, int(loop_id) if loop_id else loop_id, fn) for msg, (loop_id, fn) in msgs]
msgs.sort(key=lambda m: m[1] if m[1] else -1) # sort by loop id
for msg, loop_id, fn in msgs: # messages in log trace
if loop_id:
loop_num = max(loop_id + 1, loop_num)
if hours and loop_id == stop_li and fn == stop_fn:
break
if msg.tag and "llm" not in msg.tag and "session" not in msg.tag:
if "competition" in msg.tag:
stat[log_trace_path.name]["competition"] = msg.content
# get threshold scores
workflowexp = FBWorkspace()
if is_mle:
stdout = workflowexp.execute(
env=test_eval.env,
entry=f"mlebench grade-sample None {stat[log_trace_path.name]['competition']} --data-dir /mle/data",
)
grade_output = extract_json(stdout)
if grade_output:
bronze_threshold = grade_output["bronze_threshold"]
silver_threshold = grade_output["silver_threshold"]
gold_threshold = grade_output["gold_threshold"]
median_threshold = grade_output["median_threshold"]
if "running" in msg.tag:
if isinstance(msg.content, DSExperiment):
if msg.content.result is not None:
valid_scores[loop_id] = msg.content.result
elif "mle_score" in msg.tag:
grade_output = extract_json(msg.content)
if grade_output:
if grade_output["submission_exists"]:
made_submission_num += 1
if grade_output["score"] is not None:
test_scores[loop_id] = grade_output["score"]
# if is_mle:
# _, test_ranks[loop_id] = score_rank(
# stat[log_trace_path.name]["competition"], grade_output["score"]
# )
if grade_output["valid_submission"]:
valid_submission_num += 1
if grade_output["above_median"]:
above_median_num += 1
if grade_output["any_medal"]:
get_medal_num += 1
if grade_output["bronze_medal"]:
bronze_num += 1
if grade_output["silver_medal"]:
silver_num += 1
if grade_output["gold_medal"]:
gold_num += 1
if "feedback" in msg.tag and "evolving" not in msg.tag:
if isinstance(msg.content, ExperimentFeedback) and bool(msg.content):
success_loop_num += 1
if grade_output: # sota exp's grade output
if grade_output["gold_medal"]:
sota_exp_stat = "gold"
elif grade_output["silver_medal"]:
sota_exp_stat = "silver"
elif grade_output["bronze_medal"]:
sota_exp_stat = "bronze"
elif grade_output["above_median"]:
sota_exp_stat = "above_median"
elif grade_output["valid_submission"]:
sota_exp_stat = "valid_submission"
elif grade_output["submission_exists"]:
sota_exp_stat = "made_submission"
if grade_output["score"] is not None:
sota_exp_score = grade_output["score"]
# if is_mle:
# _, sota_exp_rank = score_rank(
# stat[log_trace_path.name]["competition"], grade_output["score"]
# )
stat[log_trace_path.name].update(
{
"loop_num": loop_num,
"made_submission_num": made_submission_num,
"valid_submission_num": valid_submission_num,
"above_median_num": above_median_num,
"get_medal_num": get_medal_num,
"bronze_num": bronze_num,
"silver_num": silver_num,
"gold_num": gold_num,
"test_scores": test_scores,
# "test_ranks": test_ranks,
"valid_scores": valid_scores,
"success_loop_num": success_loop_num,
"sota_exp_stat": sota_exp_stat,
"sota_exp_score": sota_exp_score,
# "sota_exp_rank": sota_exp_rank,
"bronze_threshold": bronze_threshold,
"silver_threshold": silver_threshold,
"gold_threshold": gold_threshold,
"median_threshold": median_threshold,
}
)
# Save the summary
save_name = f"summary_{hours}h.pkl" if hours else "summary.pkl"
save_p = log_folder / save_name
if save_p.exists():
save_p.unlink()
print(f"Old {save_name} removed.")
pd.to_pickle(stat, save_p)
# {
# "competition_id": "stanford-covid-vaccine",
# "score": null,
# "gold_threshold": 0.34728,
# "silver_threshold": 0.35175,
# "bronze_threshold": 0.3534,
# "median_threshold": 0.363095,
# "any_medal": false,
# "gold_medal": false,
# "silver_medal": false,
# "bronze_medal": false,
# "above_median": false,
# "submission_exists": true,
# "valid_submission": false,
# "is_lower_better": true,
# "created_at": "2025-01-21T11:59:33.788201",
# "submission_path": "submission.csv"
# }
def grade_summary(log_folder: str) -> None:
"""
Generate test scores for log traces in the log folder and save the summary.
"""
log_folder = Path(log_folder)
save_all_grade_info(log_folder)
summarize_folder(log_folder)
if __name__ == "__main__":
fire.Fire(
{
"grade": save_all_grade_info,
"summary": summarize_folder,
"grade_summary": grade_summary,
}
)
rdagent/log/server/README.md
-253
View File
@@ -1,253 +0,0 @@
# API
## A. Controls
### 1. /upload [POST]
#### Request
- "scenario": one of six values
1. "Finance Data Building"
2. "Finance Data Building (Reports)"
3. "Finance Model Implementation"
4. "General Model Implementation"
5. "Medical Model Implementation"
6. "Data Science"
- "files": **2** scenarios need this
1. in "Finance Data Building (Reports)" Scenario, one or more pdf files.
2. in "General Model Implementation" Scenario, one pdf file or one pdf link like `https://arxiv.org/pdf/2210.09789`
- "competition": **Data Science** Scenario need this, one of 75 competitions.
- "loops": Number of loops after which RD-Agent will automatically stop (optional; if not set, it will not stop automatically and must be stopped manually).
- "all_duration": Total duration (in hours) for which the RD-Agent should run before stopping automatically. If not set, the agent will continue running until stopped manually or by the "loops" parameter.
#### Response
- "id": a unique identifier string, such as `/home/rdagent_log/data_science/competition_A/trace_1` or `/home/rdagent_log/finance/trace_1`, used to mark the series of logs generated by this RD-Agent run.
### 2. /control [POST]
#### Request
- "id": identifier
- "action": one of three values
1. "pause"
2. "resume"
3. "stop"
#### Response
- "status": "success" / "error: ..."
### 3. /trace [POST]
Returns the sequence of Messages generated for the current id on the backend that **have not yet been returned to the frontend**.
#### Request
- "id": identifier
- "all": True / False. True means all Messages not yet provided to the frontend will be returned; False returns a random 1 to 10 Messages. In most cases, this should be True.
- "reset": True / False. Reset means the pointer for "not yet returned to the frontend" will be set back to the first Message generated for this id, i.e., return from the beginning. In most cases, this should be False.
#### Response
- a list of [Messages](#b-messages)
## B. Messages
### Research
Only **2** Message in one loop
1. hypothesis
```json
{
"tag": "research.hypothesis",
"timestamp": "<isoformat>",
"loop_id": "1",
"content": {
"hypothesis": "...",
"reason": "...",
"component": "...", // only exists in Data Science Scenario
"concise_reason": "...",
"concise_justification": "...",
"concise_observation": "...",
"concise_knowledge": "...",
}
}
```
2. tasks
```json
{
"tag": "research.tasks",
"timestamp": "<isoformat>",
"loop_id": "1",
"content": [ // list of tasks
{
"name": "...",
"description": "...",
"model_type": "...", // only exists in "Finance Model Implementation", "General Model Implementation", "Medical Model Implementation", or some tasks of "Data Science"
"architecture": "...", // same as above
"hyperparameters": "...", // same as above
},
{
}
//... same as above
]
}
```
### evolving
- 1 to 10 pairs of Messages (codes & feedbacks), each identified by an "evo_id" indicating the evolving round.
- In the **Data Science** scenario, each evolving round contains only **one task**, but the "codes" for that task may include **multiple code files**.
- In other scenarios, each evolving round may contain **multiple tasks**, but each task's "codes" will include only **one code file**.
1. codes
```json
{
"tag": "evolving.codes",
"timestamp": "<isoformat>",
"loop_id": "1",
"evo_id": "0",
"content": [ // list of task_name & codes
{
"evo_id": "0",
"target_task_name": "task_1",
"workspace": { // one or more codes
"a.py": "...<python codes>",
"b.py": "...<python codes>",
//...
}
},
{
"evo_id": "0",
"target_task_name": "task_2",
"workspace": {
"a.py": "...<python codes>",
//...
}
}
//... same as above
]
}
{
"tag": "evolving.codes",
"timestamp": "<isoformat>",
"loop_id": "1",
"evo_id": "1",
"content": [
//... same as above
]
}
```
2. feedbacks
```json
{
"tag": "evolving.feedbacks",
"timestamp": "<isoformat>",
"loop_id": "1",
"evo_id": "0",
"content": [ // list of feedbacks
{
"evo_id": "0",
"final_decision": "True", // True or False
"execution": "...",
"code": "...",
"return_checking": "..."
},
//... same as above
]
}
{
"tag": "evolving.codes",
"timestamp": "<isoformat>",
"loop_id": "1",
"evo_id": "1",
"content": [
//... same as above
]
}
```
### feedback
Each tag below appears only once per loop.
1. config (only exists in "Finance Data Building"/"Finance Data Building (Reports)"/"Finance Model Implementation")
```json
{
"tag": "feedback.config",
"timestamp": "<isoformat>",
"loop_id": "1",
"content": {
"config": "a markdown string",
}
}
```
2. return_chart (only exists in "Finance Data Building"/"Finance Data Building (Reports)"/"Finance Model Implementation")
```json
{
"tag": "feedback.return_chart",
"timestamp": "<isoformat>",
"loop_id": "1",
"content": {
"chart_html": "chart html codes string",
}
}
```
3. metric
```json
{
"tag": "feedback.metric",
"timestamp": "<isoformat>",
"loop_id": "1",
"content": {
"result": "{ \"<metric_name>\": <value>, ... }" // A JSON string containing metric names and their corresponding values.
}
}
```
4. hypothesis_feedback
```json
{
"tag": "feedback.hypothesis_feedback",
"timestamp": "<isoformat>",
"loop_id": "1",
"content": {
"decision": "True",
"reason": "...",
"exception": "...",
"observations": "...", // may not exists
"hypothesis_evaluation": "...", // may not existsc
"new_hypothesis": "...", // may not exists
}
}
```
# TODO
## Session
- How to continue.
- show & copy trace_id(name)?
-
## Page
1. remove Medical, add Finance Whole Pipeline
2.
rdagent/log/server/app.py
-562
View File
@@ -1,562 +0,0 @@
import logging
import os
import random
import traceback
from collections import defaultdict
from contextlib import redirect_stderr, redirect_stdout
from datetime import datetime, timezone
from multiprocessing import Process, Queue
from pathlib import Path
from queue import Empty
import randomname
import typer
from flask import Flask, jsonify, request, send_file, send_from_directory
from flask_cors import CORS
from werkzeug.utils import secure_filename
from rdagent.log.storage import FileStorage
from rdagent.log.ui.conf import UI_SETTING
from rdagent.log.ui.storage import WebStorage
app = Flask(__name__, static_folder=str(Path(UI_SETTING.static_path).resolve()))
CORS(app)
app.config["UI_SERVER_PORT"] = 19899
_YELLOW = "\033[33m"
_RESET = "\033[0m"
class _YellowWarningFormatter(logging.Formatter):
def format(self, record: logging.LogRecord) -> str:
if record.levelno == logging.WARNING:
record.levelname = f"{_YELLOW}{record.levelname}{_RESET}"
return super().format(record)
def _configure_app_logger() -> None:
formatter = _YellowWarningFormatter(
fmt="[%(asctime)s] %(levelname)s in %(module)s: %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
)
for handler in app.logger.handlers:
handler.setFormatter(formatter)
_configure_app_logger()
_TARGETS_WITHOUT_USER_INTERACTION = {"general_model", "fin_factor_report"}
class RDAgentTask:
def __init__(
self,
target_name: str,
kwargs: dict,
stdout_path: str,
log_trace_path: str,
scenario: str,
trace_name: str,
ui_server_port: int | None = None,
create_process: bool = True,
) -> None:
self.target_name = target_name
self.kwargs = kwargs
self.stdout_path = stdout_path
self.log_trace_path = log_trace_path
self.scenario = scenario
self.trace_name = trace_name
self.ui_server_port = ui_server_port
self.process: Process | None = None
# Two IPC queues for user interaction.
# - `user_request_q`: rdagent subprocess -> server (dicts to render on frontend)
# - `user_response_q`: server -> rdagent subprocess (user input dicts)
# NOTE: Use multiprocessing.Queue because rdagent is started as a separate process.
self.user_request_q: Queue = Queue(maxsize=1024)
self.user_response_q: Queue = Queue(maxsize=1024)
if create_process:
self.process = Process(
target=self._run,
name=f"rdagent:{self.scenario}:{self.trace_name}",
)
self.messages: list[dict] = []
self.pointers: defaultdict[str, int] = defaultdict(int)
def start(self) -> None:
if self.process is not None:
self.process.start()
def is_alive(self) -> bool:
return self.process is not None and self.process.is_alive()
def get_end_code(self) -> int:
if self.process is None or self.process.exitcode is None:
return 0
return self.process.exitcode
def stop(self) -> None:
if self.process is not None and self.process.is_alive():
self.process.terminate()
self.process.join()
# Best-effort cleanup for IPC queues.
for q in (self.user_request_q, self.user_response_q):
try:
q.cancel_join_thread()
except Exception:
pass
try:
q.close()
except Exception:
pass
def _run(self) -> None:
from rdagent.log.conf import LOG_SETTINGS
LOG_SETTINGS.set_ui_server_port(self.ui_server_port)
from rdagent.log import rdagent_logger
rdagent_logger.refresh_storages_from_settings()
rdagent_logger.set_storages_path(self.log_trace_path)
Path(self.stdout_path).parent.mkdir(parents=True, exist_ok=True)
with open(self.stdout_path, "w") as log_file:
with redirect_stdout(log_file), redirect_stderr(log_file):
rdagent_logger.rebind_console_to_current_streams()
try:
# Only interactive targets should receive IPC queues.
if self.target_name not in _TARGETS_WITHOUT_USER_INTERACTION:
self.kwargs.setdefault(
"user_interaction_queues",
(self.user_request_q, self.user_response_q),
)
if self.target_name == "data_science":
from rdagent.app.data_science.loop import main as data_science
data_science(**self.kwargs)
elif self.target_name == "general_model":
from rdagent.app.general_model.general_model import (
extract_models_and_implement as general_model,
)
general_model(**self.kwargs)
elif self.target_name == "fin_factor":
from rdagent.app.qlib_rd_loop.factor import main as fin_factor
fin_factor(**self.kwargs)
elif self.target_name == "fin_factor_report":
from rdagent.app.qlib_rd_loop.factor_from_report import (
main as fin_factor_report,
)
fin_factor_report(**self.kwargs)
elif self.target_name == "fin_model":
from rdagent.app.qlib_rd_loop.model import main as fin_model
fin_model(**self.kwargs)
elif self.target_name == "fin_quant":
from rdagent.app.qlib_rd_loop.quant import main as fin_quant
fin_quant(**self.kwargs)
else:
raise ValueError(f"Unknown target: {self.target_name}")
except Exception:
traceback.print_exc()
rdagent_processes: dict[str, RDAgentTask] = {}
log_folder_path = Path(UI_SETTING.trace_folder).absolute()
def _drain_user_requests_into_messages(task: RDAgentTask) -> None:
"""Move a single pending user-interaction request into `task.messages`.
Assumption: each rdagent process only has one active request at a time.
"""
try:
req = task.user_request_q.get_nowait()
except Empty:
return
except Exception:
return
# Standardize the message shape for the frontend.
# The agent can send either a full message dict, or a raw content dict.
if isinstance(req, dict) and {"tag", "timestamp", "content"}.issubset(req.keys()):
msg = req
else:
msg = {
"tag": "user_interaction.request",
"timestamp": datetime.now(timezone.utc).isoformat(),
"content": req,
}
task.messages.append(msg)
@app.route("/favicon.ico")
def favicon():
return send_from_directory(app.static_folder, "favicon.ico", mimetype="image/vnd.microsoft.icon")
def _normalize_static_request_path(fn: str) -> str:
static_prefix = UI_SETTING.static_path.strip("./")
if static_prefix and fn.startswith(f"{static_prefix}/"):
return fn[len(static_prefix) + 1 :]
return fn
def _get_or_create_task(trace_id: str) -> RDAgentTask:
task = rdagent_processes.get(trace_id)
if task is None:
task = RDAgentTask(
target_name="",
kwargs={},
stdout_path="",
log_trace_path=trace_id,
scenario="",
trace_name="",
ui_server_port=None,
create_process=False,
)
rdagent_processes[trace_id] = task
return task
def _resolve_stdout_path(trace_id: str) -> Path | None:
normalized_trace_id = str(trace_id or "").strip()
if not normalized_trace_id:
return None
task = rdagent_processes.get(str(log_folder_path / normalized_trace_id))
if task is None or not task.stdout_path:
return None
stdout_path = Path(task.stdout_path).resolve()
try:
if os.path.commonpath([str(stdout_path), str(log_folder_path)]) != str(log_folder_path):
return None
except ValueError:
return None
return stdout_path
def read_trace(log_path: Path, id: str = "") -> None:
fs = FileStorage(log_path)
ws = WebStorage(port=1, path=log_path)
task = _get_or_create_task(id)
task.messages = []
last_timestamp = None
for msg in fs.iter_msg():
data = ws._obj_to_json(obj=msg.content, tag=msg.tag, id=id, timestamp=msg.timestamp.isoformat())
if data:
if isinstance(data, list):
for d in data:
task.messages.append(d["msg"])
last_timestamp = msg.timestamp
else:
task.messages.append(data["msg"])
last_timestamp = msg.timestamp
now = datetime.now(timezone.utc)
if last_timestamp and (now - last_timestamp).total_seconds() > 1800:
task.messages.append(
{
"tag": "END",
"timestamp": now.isoformat(),
"content": {"error_msg": "Trace session has ended.", "end_code": 0},
}
)
# load all traces from the log folder
# for p in log_folder_path.glob("*/*/"):
# read_trace(p, id=str(p))
@app.route("/trace", methods=["POST"])
def update_trace():
data = request.get_json()
trace_id = data.get("id")
return_all = data.get("all")
reset = data.get("reset")
msg_num = random.randint(1, 10)
app.logger.info(data)
log_folder_path = Path(UI_SETTING.trace_folder).absolute()
if not trace_id:
return jsonify({"error": "Trace ID is required"}), 400
trace_id = str(log_folder_path / trace_id)
task = _get_or_create_task(trace_id)
# Make sure any pending user-interaction requests are visible to the frontend.
_drain_user_requests_into_messages(task)
if task.process is not None and not task.is_alive():
if not task.messages or task.messages[-1].get("tag") != "END":
task.messages.append(
{
"tag": "END",
"timestamp": datetime.now(timezone.utc).isoformat(),
"content": {
"error_msg": "RD-Agent process has completed.",
"end_code": task.get_end_code(),
},
}
)
app.logger.warning(f"Process for {trace_id} has ended.")
user_ip = request.remote_addr
if reset:
task.pointers[user_ip] = 0
start_pointer = task.pointers[user_ip]
end_pointer = start_pointer + msg_num
if end_pointer > len(task.messages) or return_all:
end_pointer = len(task.messages)
returned_msgs = task.messages[start_pointer:end_pointer]
task.pointers[user_ip] = end_pointer
if returned_msgs:
app.logger.info([msg["tag"] for msg in returned_msgs])
return jsonify(returned_msgs), 200
@app.route("/stdout", methods=["GET"])
def download_stdout_file():
trace_id = request.args.get("id", "")
stdout_path = _resolve_stdout_path(trace_id)
if stdout_path is None:
return jsonify({"error": "Trace ID is required or invalid"}), 400
if not stdout_path.exists() or not stdout_path.is_file():
return jsonify({"error": "Stdout file not found"}), 404
return send_file(
stdout_path,
as_attachment=True,
download_name=stdout_path.name,
mimetype="text/plain",
)
@app.route("/upload", methods=["POST"])
def upload_file():
# 获取请求体中的字段
global rdagent_processes
scenario = request.form.get("scenario")
files = request.files.getlist("files")
competition = request.form.get("competition")
loop_n = request.form.get("loops")
all_duration = request.form.get("all_duration")
# scenario = "Data Science Loop"
if scenario == "Data Science":
competition = competition[10:] # Eg. MLE-Bench:aerial-cactus-competition
trace_name = f"{competition}-{randomname.get_name()}"
else:
trace_name = randomname.get_name()
trace_files_path = log_folder_path / "uploads" / scenario / trace_name
log_trace_path = (log_folder_path / scenario / trace_name).absolute()
stdout_path = log_folder_path / scenario / f"{trace_name}.log"
if not stdout_path.exists():
stdout_path.parent.mkdir(parents=True, exist_ok=True)
# save files
for file in files:
if file:
p = (log_folder_path / "uploads" / scenario / trace_name).resolve()
sanitized_filename = secure_filename(file.filename) # Sanitize filename
target_path = (p / sanitized_filename).resolve() # Normalize target path
# Ensure target_path is within the allowed base directory
if os.path.commonpath([str(target_path), str(p)]) == str(p) and target_path.is_file() == False:
if not p.exists():
p.mkdir(parents=True, exist_ok=True)
file.save(target_path)
else:
return jsonify({"error": "Invalid file path"}), 400
target_name = None
kwargs = {}
loop_n_val = int(loop_n) if loop_n else None
all_duration_val = f"{all_duration}h" if all_duration else None
if scenario == "Finance Data Building":
target_name = "fin_factor"
kwargs = {
"loop_n": loop_n_val,
"all_duration": all_duration_val,
"base_features_path": str(trace_files_path),
}
if scenario == "Finance Model Implementation":
target_name = "fin_model"
kwargs = {
"loop_n": loop_n_val,
"all_duration": all_duration_val,
"base_features_path": str(trace_files_path),
}
if scenario == "Finance Whole Pipeline":
target_name = "fin_quant"
kwargs = {
"loop_n": loop_n_val,
"all_duration": all_duration_val,
"base_features_path": str(trace_files_path),
}
if scenario == "Finance Data Building (Reports)":
target_name = "fin_factor_report"
kwargs = {"report_folder": str(trace_files_path), "all_duration": all_duration_val}
if scenario == "General Model Implementation":
if len(files) == 0: # files is one link
rfp = request.form.get("files")[0]
else: # one file is uploaded
rfp = str(trace_files_path / files[0].filename)
target_name = "general_model"
kwargs = {"report_file_path": rfp}
if scenario == "Data Science":
target_name = "data_science"
kwargs = {"competition": competition, "loop_n": loop_n_val, "timeout": all_duration_val}
if target_name is None:
return jsonify({"error": "Unknown scenario"}), 400
app.logger.info(f"Started process for {log_trace_path} with target: {target_name}, kwargs: {kwargs}")
task = RDAgentTask(
target_name=target_name,
kwargs=kwargs,
stdout_path=str(stdout_path),
log_trace_path=str(log_trace_path),
scenario=scenario,
trace_name=trace_name,
ui_server_port=app.config["UI_SERVER_PORT"],
)
task.start()
app.logger.warning(f"Task {log_trace_path} started.")
rdagent_processes[str(log_trace_path)] = task
return (
jsonify(
{
"id": f"{scenario}/{trace_name}",
}
),
200,
)
@app.route("/receive", methods=["POST"])
def receive_msgs():
try:
data = request.get_json()
if not data:
return jsonify({"error": "No JSON data received"}), 400
except Exception as e:
return jsonify({"error": "Internal Server Error"}), 500
if isinstance(data, list):
for d in data:
task = _get_or_create_task(d["id"])
task.messages.append(d["msg"])
else:
task = _get_or_create_task(data["id"])
task.messages.append(data["msg"])
return jsonify({"status": "success"}), 200
@app.route("/user_interaction/submit", methods=["POST"])
def submit_user_interaction_response():
"""Frontend submits a user response; server forwards it to the rdagent subprocess via IPC queue."""
data = request.get_json(silent=True) or {}
trace_id = data.get("id")
payload = data.get("payload")
if not trace_id:
return jsonify({"error": "Trace ID is required"}), 400
if payload is None:
return jsonify({"error": "Missing 'payload'"}), 400
trace_id = str(log_folder_path / trace_id)
task = _get_or_create_task(trace_id)
try:
task.user_response_q.put(payload, block=False)
except Exception as e:
return jsonify({"error": f"Failed to enqueue user response: {e}"}), 500
return jsonify({"status": "success"}), 200
@app.route("/control", methods=["POST"])
def control_process():
global rdagent_processes
data = request.get_json()
app.logger.info(data)
if not data or "id" not in data or "action" not in data:
return jsonify({"error": "Missing 'id' or 'action' in request"}), 400
id = str(log_folder_path / data["id"])
action = data["action"]
if action != "stop":
return jsonify({"error": "Only 'stop' action is supported"}), 400
if id not in rdagent_processes or rdagent_processes[id] is None:
return jsonify({"error": "No running process for given id"}), 400
task = rdagent_processes[id]
if task.process is None:
return jsonify({"error": "No running process for given id"}), 400
try:
if task.is_alive():
task.stop()
if not task.messages or task.messages[-1].get("tag") != "END":
task.messages.append(
{
"tag": "END",
"timestamp": datetime.now(timezone.utc).isoformat(),
"content": {"error_msg": "RD-Agent process was stopped by user.", "end_code": -1},
}
)
app.logger.warning(f"Process for {id} has been stopped.")
return jsonify({"status": "stopped"}), 200
except Exception as e:
return jsonify({"error": f"Failed to {action} process, {e}"}), 500
@app.route("/test", methods=["GET"])
def test():
# return 'Hello, World!'
msgs = {k: [i["tag"] for i in task.messages] for k, task in rdagent_processes.items()}
pointers = {k: dict(task.pointers) for k, task in rdagent_processes.items()}
return jsonify({"msgs": msgs, "pointers": pointers}), 200
@app.route("/", methods=["GET"])
def index():
# return 'Hello, World!'
# return {k: [i["tag"] for i in v] for k, v in msgs_for_frontend.items()}
return send_from_directory(app.static_folder, "index.html")
@app.route("/<path:fn>", methods=["GET"])
def server_static_files(fn):
return send_from_directory(app.static_folder, _normalize_static_request_path(fn))
def main(port: int = 19899):
app.config["UI_SERVER_PORT"] = port
app.run(debug=False, host="0.0.0.0", port=port)
if __name__ == "__main__":
typer.run(main)
rdagent/log/server/debug_app.py
-174
View File
@@ -1,174 +0,0 @@
import multiprocessing
import os
import random
import signal
import subprocess
import threading
import time
from collections import defaultdict
from datetime import datetime, timezone
from pathlib import Path
import randomname
import typer
from flask import Flask, jsonify, request, send_from_directory
from flask_cors import CORS
from rdagent.log.ui.conf import UI_SETTING
app = Flask(__name__, static_folder=UI_SETTING.static_path)
CORS(app)
rdagent_processes = defaultdict()
server_port = 19899
@app.route("/favicon.ico")
def favicon():
return send_from_directory(app.static_folder, "favicon.ico", mimetype="image/vnd.microsoft.icon")
msgs_for_frontend = defaultdict(list)
pointers = defaultdict(int)
@app.route("/trace", methods=["POST"])
def update_trace():
global pointers, msgs_for_frontend
data = request.get_json()
# app.logger.info(data)
trace_id = data.get("id")
return_all = data.get("all")
reset = data.get("reset")
msg_num = random.randint(1, 10)
if reset:
pointers[trace_id] = 0
end_pointer = pointers[trace_id] + msg_num
if end_pointer > len(msgs_for_frontend[trace_id]) or return_all:
end_pointer = len(msgs_for_frontend[trace_id])
returned_msgs = msgs_for_frontend[trace_id][pointers[trace_id] : end_pointer]
pointers[trace_id] = end_pointer
# if len(returned_msgs):
# app.logger.info(data)
# app.logger.info([i["tag"] for i in returned_msgs])
# try:
# import json
# resp = json.dumps(returned_msgs, ensure_ascii=False)
# except Exception as e:
# app.logger.error(f"Error in jsonify: {e}")
# for msg in returned_msgs:
# try:
# rr = json.dumps(msg, ensure_ascii=False)
# except Exception as e:
# app.logger.error(f"Error in jsonify individual message: {e}")
# app.logger.error(msg)
return jsonify(returned_msgs), 200
@app.route("/upload", methods=["POST"])
def upload_file():
# 获取请求体中的字段
global rdagent_processes, server_port, msgs_for_frontend
scenario = request.form.get("scenario")
files = request.files.getlist("files")
competition = request.form.get("competition")
loop_n = request.form.get("loops")
all_duration = request.form.get("all_duration")
log_folder_path = Path("/home/bowen/workspace/new_traces").absolute()
if scenario == "Data Science":
trace_path = log_folder_path / "o1-preview" / f"{competition[10:]}.1"
else:
trace_path = log_folder_path / scenario
id = f"{scenario}/{randomname.get_name()}"
def read_trace(log_path: Path, t: float = 0.2, id: str = "") -> None:
from rdagent.log.storage import FileStorage
from rdagent.log.ui.storage import WebStorage
fs = FileStorage(log_path)
ws = WebStorage(port=1, path=log_path)
msgs_for_frontend[id] = []
for msg in fs.iter_msg():
data = ws._obj_to_json(obj=msg.content, tag=msg.tag, id=id, timestamp=msg.timestamp.isoformat())
if data:
if isinstance(data, list):
for d in data:
time.sleep(t)
msgs_for_frontend[id].append(d["msg"])
else:
time.sleep(t)
msgs_for_frontend[id].append(data["msg"])
msgs_for_frontend[id].append({"tag": "END", "timestamp": datetime.now(timezone.utc).isoformat(), "content": {}})
# 启动后台线程，不阻塞 return
threading.Thread(target=read_trace, args=(trace_path, 0.5, id), daemon=True).start()
return jsonify({"id": id}), 200
@app.route("/receive", methods=["POST"])
def receive_msgs():
try:
data = request.get_json()
app.logger.info(data["msg"]["tag"])
if not data:
return jsonify({"error": "No JSON data received"}), 400
except Exception as e:
return jsonify({"error": "Internal Server Error"}), 500
if isinstance(data, list):
for d in data:
msgs_for_frontend[d["id"]].append(d["msg"])
else:
msgs_for_frontend[data["id"]].append(data["msg"])
return jsonify({"status": "success"}), 200
@app.route("/control", methods=["POST"])
def control_process():
global rdagent_processes
data = request.get_json()
app.logger.info(data)
if not data or "id" not in data or "action" not in data:
return jsonify({"error": "Missing 'id' or 'action' in request"}), 400
id = data["id"]
action = data["action"]
return jsonify({"status": "success", "message": f"Received action '{action}' for process with id '{id}'"})
@app.route("/test", methods=["GET"])
def test():
# return 'Hello, World!'
return {k: [i["tag"] for i in v] for k, v in msgs_for_frontend.items()}
@app.route("/", methods=["GET"])
def index():
# return 'Hello, World!'
# return {k: [i["tag"] for i in v] for k, v in msgs_for_frontend.items()}
return send_from_directory(app.static_folder, "index.html")
@app.route("/<path:fn>", methods=["GET"])
def server_static_files(fn):
return send_from_directory(app.static_folder, fn)
def main(port: int = 19899):
global server_port
server_port = port
app.run(debug=True, host="0.0.0.0", port=port)
if __name__ == "__main__":
typer.run(main)
rdagent/log/storage.py
-116
View File
@@ -1,116 +0,0 @@
import json
import pickle
import re
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Generator, Literal
from .base import Message, Storage
from .utils import gen_datetime
LOG_LEVEL = Literal["DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"]
def _remove_empty_dir(path: Path) -> None:
"""
Recursively remove empty directories.
This function will remove the directory if it is empty after removing its subdirectories.
"""
if path.is_dir():
sub_dirs = [sub for sub in path.iterdir() if sub.is_dir()]
for sub in sub_dirs:
_remove_empty_dir(sub)
if not any(path.iterdir()):
path.rmdir()
class FileStorage(Storage):
"""
The info are logginged to the file systems
TODO: describe the storage format
"""
def __init__(self, path: str | Path) -> None:
self.path = Path(path)
def log(
self,
obj: object,
tag: str = "",
timestamp: datetime | None = None,
save_type: Literal["json", "text", "pkl"] = "pkl",
**kwargs: Any,
) -> str | Path:
# TODO: We can remove the timestamp after we implement PipeLog
timestamp = gen_datetime(timestamp)
cur_p = self.path / tag.replace(".", "/")
cur_p.mkdir(parents=True, exist_ok=True)
path = cur_p / f"{timestamp.strftime('%Y-%m-%d_%H-%M-%S-%f')}.log"
if save_type == "json":
path = path.with_suffix(".json")
with path.open("w") as f:
try:
json.dump(obj, f)
except TypeError:
json.dump(json.loads(str(obj)), f)
return path
elif save_type == "pkl":
path = path.with_suffix(".pkl")
with path.open("wb") as f:
pickle.dump(obj, f)
return path
elif save_type == "text":
obj = str(obj)
with path.open("w") as f:
f.write(obj)
return path
log_pattern = re.compile(
r"(?P<timestamp>\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}\.\d{3}) \| "
r"(?P<level>DEBUG|INFO|WARNING|ERROR|CRITICAL) *\| "
r"(?P<caller>.+:.+:\d+) - "
)
def iter_msg(self, tag: str | None = None, pattern: str | None = None) -> Generator[Message, None, None]:
msg_l = []
if pattern:
pkl_files = pattern
elif tag:
pkl_files = f"**/{tag.replace('.','/')}/**/*.pkl"
else:
pkl_files = "**/*.pkl"
for file in self.path.glob(pkl_files):
if file.name == "debug_llm.pkl":
continue
pkl_log_tag = ".".join(file.relative_to(self.path).as_posix().replace("/", ".").split(".")[:-3])
pid = file.parent.name
with file.open("rb") as f:
content = pickle.load(f)
timestamp = datetime.strptime(file.stem, "%Y-%m-%d_%H-%M-%S-%f").replace(tzinfo=timezone.utc)
m = Message(tag=pkl_log_tag, level="INFO", timestamp=timestamp, caller="", pid_trace=pid, content=content)
msg_l.append(m)
msg_l.sort(key=lambda x: x.timestamp)
for m in msg_l:
yield m
def truncate(self, time: datetime) -> None:
for file in self.path.glob("**/*.pkl"):
timestamp = datetime.strptime(file.stem, "%Y-%m-%d_%H-%M-%S-%f").replace(tzinfo=timezone.utc)
if timestamp > time.replace(tzinfo=timezone.utc):
file.unlink()
_remove_empty_dir(self.path)
def __str__(self) -> str:
return f"FileStorage({self.path})"
rdagent/log/timer.py
-86
View File
@@ -1,86 +0,0 @@
import re
from datetime import datetime, timedelta
from rdagent.core.utils import SingletonBaseClass
from rdagent.log import rdagent_logger as logger
class RDAgentTimer:
def __init__(self) -> None:
self.started: bool = False
self.target_time: datetime | None = None
self.all_duration: timedelta | None = None
self._remain_time_duration: timedelta | None = None
def reset(self, all_duration: str | timedelta) -> None:
if isinstance(all_duration, str):
pattern = re.compile(r"^\s*(\d*\.?\d+)\s*([smhd]?)\s*$")
match = pattern.match(all_duration)
if not match:
return None
value = float(match.group(1))
unit = match.group(2)
if unit == "s":
self.all_duration = timedelta(seconds=value)
elif unit == "m":
self.all_duration = timedelta(minutes=value)
elif unit == "h":
self.all_duration = timedelta(hours=value)
elif unit == "d":
self.all_duration = timedelta(days=value)
else:
self.all_duration = timedelta(seconds=value)
elif isinstance(all_duration, timedelta):
self.all_duration = all_duration
self.target_time = datetime.now() + self.all_duration
logger.info(f"Timer set to {self.all_duration} seconds and counting down.")
self.started = True
return None
def restart_by_remain_time(self) -> None:
if self._remain_time_duration is not None:
self.target_time = datetime.now() + self._remain_time_duration
self.started = True
logger.info(f"Timer restarted with remaining time: {self._remain_time_duration}")
else:
logger.warning("No remaining time to restart the timer.")
return None
def add_duration(self, duration: timedelta) -> None:
if self.started and self.target_time is not None:
logger.info(f"Adding {duration} to the timer. Currently {self.remain_time()} remains.")
self.target_time = self.target_time + duration
self.update_remain_time()
def is_timeout(self) -> bool:
if self.started and self.target_time is not None:
self.update_remain_time()
if datetime.now() > self.target_time:
return True
return False
def update_remain_time(self) -> None:
if self.started and self.target_time is not None:
self._remain_time_duration = self.target_time - datetime.now()
return None
def remain_time(self) -> timedelta | None:
if self.started:
self.update_remain_time()
return self._remain_time_duration
return None
class RDAgentTimerWrapper(SingletonBaseClass):
def __init__(self) -> None:
self.timer: RDAgentTimer = RDAgentTimer()
self.api_fail_count: int = 0
self.latest_api_fail_time: datetime | None = None
def replace_timer(self, timer: RDAgentTimer) -> None:
self.timer = timer
logger.info("Timer replaced successfully.")
RD_Agent_TIMER_wrapper: RDAgentTimerWrapper = RDAgentTimerWrapper()
rdagent/log/ui/__init__.py
-7
View File
@@ -1,7 +0,0 @@
"""
UI is a kind of view for user.
We are not sure how generality of the UI, we can't make decision among following options:
- in general folder like rdagent/log/ui
- It is for specific scenario rdagent/scenarios/
"""
rdagent/log/ui/aide.py
-53
View File
@@ -1,53 +0,0 @@
# %%
import json
from pathlib import Path
import streamlit as st
from rdagent.log.ui.conf import UI_SETTING
from rdagent.utils.repo.diff import generate_diff_from_dict
aide_path = UI_SETTING.aide_path
if not Path(aide_path).exists():
st.error(f"Path {aide_path} does not exist, set it by `UI_AIDE_PATH`")
st.stop()
jps = [str(i) for i in Path(aide_path).rglob("**/filtered_journal.json")]
jps = sorted(jps)
# st.write(jps)
left, right = st.columns([1, 4])
with left:
default = 0
ppp = f"{aide_path}/{st.query_params.get('jnp')}/logs/filtered_journal.json"
if ppp in jps:
default = jps.index(ppp)
jnp = st.radio("Select Journal", options=jps, index=default, format_func=lambda x: str(x).split("/")[-3])
jnp = Path(jnp)
with jnp.open("r") as f:
d = json.load(f)
# with jnp_.open("r") as f:
# d1 = json.load(f)
nm = {nd["id"]: nd for nd in d["nodes"]}
# %%
with right:
st.header("AIDE trace", divider="rainbow")
st.subheader(jnp)
for c, p in d["node2parent"].items():
f = nm[p]
t = nm[c]
df_lines = generate_diff_from_dict({"aide.py": f["code"]}, {"aide.py": t["code"]})
with st.expander(f"Node {p} -> {c}"):
st.markdown(f"## Parent ({f['metric']['value']}) Analysis")
st.code(f["analysis"], wrap_lines=True)
st.markdown(f"## Child ({t['metric']['value']}) Plan")
st.code(t["plan"], wrap_lines=True)
st.markdown("## Diff")
st.code("".join(df_lines), language="diff", wrap_lines=True)
# print("".join(df_lines))
# %%
rdagent/log/ui/app.py
-1133
View File
File diff suppressed because it is too large Load Diff
rdagent/log/ui/conf.py
-24
View File
@@ -1,24 +0,0 @@
from pydantic_settings import SettingsConfigDict
from rdagent.core.conf import ExtendedBaseSettings
class UIBasePropSetting(ExtendedBaseSettings):
model_config = SettingsConfigDict(env_prefix="UI_", protected_namespaces=())
default_log_folders: list[str] = ["./log"]
baseline_result_path: str = "./baseline.csv"
aide_path: str = "./aide"
amlt_path: str = "/data/share_folder_local/amlt"
static_path: str = "./git_ignore_folder/static"
trace_folder: str = "./git_ignore_folder/traces"
enable_cache: bool = True
UI_SETTING = UIBasePropSetting()
rdagent/log/ui/ds_summary.py
-208
View File
@@ -1,208 +0,0 @@
"""
Please refer to rdagent/log/ui/utils.py:get_summary_df for more detailed documents about metrics
"""
import re
from pathlib import Path
import pandas as pd
import plotly.express as px
import streamlit as st
from streamlit import session_state as state
from rdagent.log.ui.utils import (
ALL,
HIGH,
LITE,
MEDIUM,
curve_figure,
get_statistics_df,
get_summary_df,
lite_curve_figure,
percent_df,
)
from rdagent.scenarios.kaggle.kaggle_crawler import get_metric_direction
def curves_win(summary: dict):
# draw curves
cbwin1, cbwin2 = st.columns(2)
if cbwin1.toggle("Show Curves", key="show_curves"):
for k, v in summary.items():
with st.container(border=True):
st.markdown(f"**:blue[{k}] - :violet[{v['competition']}]**")
try:
tscores = {k: v for k, v in v["test_scores"].items()}
tscores = pd.Series(tscores)
vscores = {}
for k, vs in v["valid_scores"].items():
if not vs.index.is_unique:
st.warning(
f"Loop {k}'s valid scores index are not unique, only the last one will be kept to show."
)
st.write(vs)
vscores[k] = vs[~vs.index.duplicated(keep="last")].iloc[:, 0]
if len(vscores) > 0:
metric_name = list(vscores.values())[0].name
else:
metric_name = "None"
vscores = pd.DataFrame(vscores)
if "ensemble" in vscores.index:
ensemble_row = vscores.loc[["ensemble"]]
vscores = pd.concat([ensemble_row, vscores.drop("ensemble")])
vscores = vscores.T
vscores["test"] = tscores
vscores.index = [f"L{i}" for i in vscores.index]
vscores.columns.name = metric_name
st.plotly_chart(curve_figure(vscores))
except Exception as e:
import traceback
st.markdown("- Error: " + str(e))
st.code(traceback.format_exc())
st.markdown("- Valid Scores: ")
# st.write({k: type(v) for k, v in v["valid_scores"].items()})
st.json(v["valid_scores"])
if cbwin2.toggle("Show Curves (Lite)", key="show_curves_lite"):
st.pyplot(lite_curve_figure(summary))
def all_summarize_win():
def shorten_folder_name(folder: str) -> str:
if "amlt" in folder:
return folder[folder.rfind("amlt") + 5 :].split("/")[0]
if "ep" in folder:
return folder[folder.rfind("ep") :]
return folder
selected_folders = st.multiselect(
"Show these folders",
state.log_folders,
state.log_folders,
format_func=shorten_folder_name,
)
for lf in selected_folders:
if not (Path(lf) / "summary.pkl").exists():
st.warning(
f"summary.pkl not found in **{lf}**\n\nRun:`dotenv run -- python rdagent/log/mle_summary.py grade_summary --log_folder={lf} --hours=<>`"
)
summary = {}
dfs = []
for lf in selected_folders:
s, df = get_summary_df(lf)
df.index = [f"{shorten_folder_name(lf)} - {idx}" for idx in df.index]
dfs.append(df)
summary.update({f"{shorten_folder_name(lf)} - {k}": v for k, v in s.items()})
base_df = pd.concat(dfs)
valid_rate = float(base_df.get("Valid Improve", pd.Series()).mean())
test_rate = float(base_df.get("Test Improve", pd.Series()).mean())
submit_merge_rate = float(base_df.get("Submit Merge", pd.Series()).mean())
merge_sota_avg = float(base_df.get("Merge Sota", pd.Series()).mean())
base_df = percent_df(base_df)
base_df.insert(0, "Select", True)
bt1, bt2 = st.columns(2)
select_lite_level = bt2.selectbox(
"Select MLE-Bench Competitions Level",
options=["ALL", "HIGH", "MEDIUM", "LITE"],
index=0,
key="select_lite_level",
)
if select_lite_level != "ALL":
if select_lite_level == "HIGH":
lite_set = set(HIGH)
elif select_lite_level == "MEDIUM":
lite_set = set(MEDIUM)
elif select_lite_level == "LITE":
lite_set = set(LITE)
else:
lite_set = set()
base_df["Select"] = base_df["Competition"].isin(lite_set)
else:
base_df["Select"] = True # select all if ALL is chosen
if bt1.toggle("Select Best", key="select_best"):
def apply_func(cdf: pd.DataFrame):
cp = base_df.loc[cdf.index[0], "Competition"]
md = get_metric_direction(cp)
# If SOTA Exp Score (valid, to_submit) column is empty, return the first index
if cdf["SOTA Exp Score (valid, to_submit)"].dropna().empty:
return cdf.index[0]
if md:
best_idx = cdf["SOTA Exp Score (valid, to_submit)"].idxmax()
else:
best_idx = cdf["SOTA Exp Score (valid, to_submit)"].idxmin()
return best_idx
best_idxs = base_df.groupby("Competition").apply(apply_func, include_groups=False)
base_df["Select"] = base_df.index.isin(best_idxs.values)
base_df = st.data_editor(
base_df,
column_config={
"Select": st.column_config.CheckboxColumn("Select", help="Stat this trace.", disabled=False),
},
disabled=(col for col in base_df.columns if col not in ["Select"]),
)
st.markdown("Ours vs Base: `math.exp(abs(math.log(sota_exp_score / baseline_score)))`")
# 统计选择的比赛
base_df = base_df[base_df["Select"]]
st.markdown(f"**统计的比赛数目: :red[{base_df.shape[0]}]**")
stat_win_left, stat_win_right = st.columns(2)
with stat_win_left:
stat_df = get_statistics_df(base_df)
st.dataframe(stat_df.round(2))
markdown_table = f"""
| xxx | {stat_df.iloc[0,1]:.1f} | {stat_df.iloc[1,1]:.1f} | {stat_df.iloc[2,1]:.1f} | {stat_df.iloc[3,1]:.1f} | {stat_df.iloc[4,1]:.1f} | {stat_df.iloc[5,1]:.1f} | {stat_df.iloc[6,1]:.1f} |
| Valid Improve {valid_rate * 100:.2f}% | Test Improve {test_rate * 100:.2f}% | Submit Merge {submit_merge_rate * 100:.2f}% | Merge Sota {merge_sota_avg * 100:.2f}% |
"""
st.text(markdown_table)
with stat_win_right:
Loop_counts = base_df["Total Loops"]
# Create histogram
fig = px.histogram(
Loop_counts, nbins=15, title="Distribution of Total Loops", color_discrete_sequence=["#3498db"]
)
fig.update_layout(title_font_size=16, title_font_color="#2c3e50")
# Calculate statistics
mean_value = Loop_counts.mean()
median_value = Loop_counts.median()
# Add mean and median lines
fig.add_vline(x=mean_value, line_color="#e74c3c", line_width=3)
fig.add_vline(x=median_value, line_color="#f39c12", line_width=3)
fig.add_annotation(
x=0.02,
y=0.95,
xref="paper",
yref="paper",
text=f"<span style='color:#e74c3c; font-weight:bold'>Mean: {mean_value:.1f}</span><br><span style='color:#f39c12; font-weight:bold'>Median: {median_value:.1f}</span>",
showarrow=False,
bgcolor="rgba(255,255,255,0.9)",
bordercolor="rgba(128,128,128,0.5)",
borderwidth=1,
font=dict(size=12, color="#333333"),
)
st.plotly_chart(fig, use_container_width=True)
# write curve
st.subheader("Curves", divider="rainbow")
curves_win(summary)
with st.container(border=True):
try:
all_summarize_win()
except Exception as e:
import traceback
st.error(f"Error occurred when show summary:\n{e}")
st.code(traceback.format_exc())
rdagent/log/ui/ds_trace.py
-1205
View File
File diff suppressed because it is too large Load Diff
rdagent/log/ui/ds_user_interact.py
-172
View File
@@ -1,172 +0,0 @@
import json
import pickle
import time
from datetime import datetime, timedelta
from pathlib import Path
import streamlit as st
from streamlit import session_state as state
from rdagent.app.data_science.conf import DS_RD_SETTING
st.set_page_config(layout="wide", page_title="RD-Agent_user_interact", page_icon="🎓", initial_sidebar_state="expanded")
# 初始化session state
if "sessions" not in state:
state.sessions = {}
if "selected_session_name" not in state:
state.selected_session_name = None
def render_main_content():
"""渲染主要内容区域"""
if state.selected_session_name is not None and state.selected_session_name in state.sessions:
selected_session_data = state.sessions[state.selected_session_name]
if selected_session_data is not None:
st.title(
f"Session: {state.selected_session_name[:4]} with competition {selected_session_data['competition']}"
)
st.title("Contextual Information:")
st.subheader("Competition scenario:", divider=True)
scenario = st.code(selected_session_data["scenario_description"], language="yaml")
st.subheader("Former attempts summary:", divider=True)
scenario = st.code(selected_session_data["ds_trace_desc"], language="yaml")
if selected_session_data["current_code"] != "":
st.subheader("Current SOTA code", divider=True)
scenario = st.code(
body=selected_session_data["current_code"],
language="python",
)
st.subheader("Hypothesis candidates:", divider=True)
hypothesis_candidates = selected_session_data["hypothesis_candidates"]
tabs = st.tabs(
[
f"{'✅' if i == selected_session_data['target_hypothesis_index'] or selected_session_data['target_hypothesis_index'] == -1 else ''}Hypothesis {i+1}"
for i in range(len(hypothesis_candidates))
]
)
for index, hypothesis in enumerate(hypothesis_candidates):
with tabs[index]:
st.code(str(hypothesis), language="yaml")
st.text("✅ means picked as target hypothesis")
st.title("Decisions to make:")
with st.form(key="user_form"):
st.caption("Please modify the fields below and submit to provide your feedback.")
target_hypothesis = st.text_area(
"Target hypothesis: (you can copy from candidates)",
value=(original_hypothesis := selected_session_data["target_hypothesis"].hypothesis),
height="content",
)
target_task = st.text_area(
"Target task description:",
value=(original_task_desc := selected_session_data["task"].description),
height="content",
)
original_user_instruction = selected_session_data.get("user_instruction")
user_instruction_list = []
if selected_session_data.get("former_user_instructions") is not None:
st.caption(
"Former user instructions, you can modify or delete the content to remove certain instruction."
)
for user_instruction in selected_session_data.get("former_user_instructions"):
user_instruction_list.append(
st.text_area("Former user instruction", value=user_instruction, height="content")
)
user_instruction_list.append(st.text_area("Add new user instruction", value="", height="content"))
submit = st.form_submit_button("Submit")
approve = st.form_submit_button("Approve without changes")
if submit or approve:
if approve:
submit_dict = {
"action": "confirm",
}
else:
user_instruction_str_list = [ui for ui in user_instruction_list if ui.strip() != ""]
user_instruction_str_list = (
None if len(user_instruction_str_list) == 0 else user_instruction_str_list
)
action = (
"confirm"
if target_hypothesis == original_hypothesis
and target_task == original_task_desc
and user_instruction_str_list == original_user_instruction
else "rewrite"
)
submit_dict = {
"target_hypothesis": target_hypothesis,
"task_description": target_task,
"user_instruction": user_instruction_str_list,
"action": action,
}
json.dump(
submit_dict,
open(
DS_RD_SETTING.user_interaction_mid_folder / f"{state.selected_session_name}_RET.json", "w"
),
)
Path(DS_RD_SETTING.user_interaction_mid_folder / f"{state.selected_session_name}.pkl").unlink(
missing_ok=True
)
st.success("Your feedback has been submitted. Thank you!")
time.sleep(5)
state.selected_session_name = None
if st.button("Extend expiration by 60s"):
session_data = pickle.load(
open(DS_RD_SETTING.user_interaction_mid_folder / f"{state.selected_session_name}.pkl", "rb")
)
session_data["expired_datetime"] = session_data["expired_datetime"] + timedelta(seconds=60)
pickle.dump(
session_data,
open(DS_RD_SETTING.user_interaction_mid_folder / f"{state.selected_session_name}.pkl", "wb"),
)
else:
st.warning("Please select a session from the sidebar.")
# 每秒更新一次sessions
@st.fragment(run_every=1)
def update_sessions():
log_folder = Path(DS_RD_SETTING.user_interaction_mid_folder)
state.sessions = {}
for session_file in log_folder.glob("*.pkl"):
try:
session_data = pickle.load(open(session_file, "rb"))
if session_data["expired_datetime"] > datetime.now():
state.sessions[session_file.stem] = session_data
else:
session_file.unlink(missing_ok=True)
ret_file = log_folder / f"{session_file.stem}_RET.json"
ret_file.unlink(missing_ok=True)
except Exception as e:
continue
render_main_content()
@st.fragment(run_every=1)
def render_sidebar():
st.title("R&D-Agent User Interaction Portal")
if state.sessions:
st.header("Active Sessions")
st.caption("Click a session to view:")
session_names = [name for name in state.sessions]
for session_name in session_names:
with st.container(border=True):
remaining = state.sessions[session_name]["expired_datetime"] - datetime.now()
total_sec = int(remaining.total_seconds())
label = f"{total_sec}s to expire" if total_sec > 0 else "Expired"
if st.button(f"session id:{session_name[:4]}", key=f"session_btn_{session_name}"):
state.selected_session_name = session_name
state.data = state.sessions[session_name]
st.markdown(f"⏳ {label}")
else:
st.warning("No active sessions available. Please wait.")
update_sessions()
with st.sidebar:
render_sidebar()
rdagent/log/ui/dsapp.py
-51
View File
@@ -1,51 +0,0 @@
from pathlib import Path
import streamlit as st
from streamlit import session_state as state
from rdagent.app.data_science.loop import DataScienceRDLoop
from rdagent.log.ui.conf import UI_SETTING
def convert_log_folder_str(lf: str) -> str:
if "/" not in lf:
return f"{UI_SETTING.amlt_path}/{lf.strip()}/combined_logs"
return lf.strip()
def extract_amlt_name(x: str) -> str:
if "amlt" not in x:
return x
return x[x.rfind("amlt") + 5 :].split("/")[0]
# 设置主日志路径
if "log_folder" not in state:
state.log_folder = Path("./log")
if "log_folders" not in state:
state.log_folders = [convert_log_folder_str(i) for i in UI_SETTING.default_log_folders]
summary_page = st.Page("ds_summary.py", title="Summary", icon="📊")
trace_page = st.Page("ds_trace.py", title="Trace", icon="📈")
aide_page = st.Page("aide.py", title="Aide", icon="🧑‍🏫")
st.set_page_config(layout="wide", page_title="RD-Agent", page_icon="🎓", initial_sidebar_state="expanded")
st.navigation([summary_page, trace_page, aide_page]).run()
# UI - Sidebar
with st.sidebar:
st.subheader("Pages", divider="rainbow")
st.page_link(summary_page, icon="📊")
st.page_link(trace_page, icon="📈")
st.page_link(aide_page, icon="🧑‍🏫")
st.subheader("Settings", divider="rainbow")
with st.form("log_folder_form", border=False):
log_folder_str = st.text_area(
"**Log Folders**(split by ';')", value=";".join(extract_amlt_name(i) for i in state.log_folders)
)
if st.form_submit_button("Confirm"):
state.log_folders = [
convert_log_folder_str(folder) for folder in log_folder_str.split(";") if folder.strip()
]
st.rerun()
rdagent/log/ui/flow.png
BIN
View File
Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 123 KiB

rdagent/log/ui/llm_st.py
-306
View File
@@ -1,306 +0,0 @@
import argparse
import json
import pickle
import re
import time
from pathlib import Path
import streamlit as st
from streamlit import session_state
from rdagent.log.ui.conf import UI_SETTING
from rdagent.log.utils import extract_evoid, extract_loopid_func_name
st.set_page_config(layout="wide", page_title="debug_llm", page_icon="🎓", initial_sidebar_state="expanded")
# 获取 log_path 参数
parser = argparse.ArgumentParser(description="RD-Agent Streamlit App")
parser.add_argument("--log_dir", type=str, help="Path to the log directory")
args = parser.parse_args()
def get_folders_sorted(log_path):
"""缓存并返回排序后的文件夹列表，并加入进度打印"""
with st.spinner("正在加载文件夹列表..."):
folders = sorted(
(folder for folder in log_path.iterdir() if folder.is_dir() and list(folder.iterdir())),
key=lambda folder: folder.stat().st_mtime,
reverse=True,
)
st.write(f"找到 {len(folders)} 个文件夹")
return [folder.name for folder in folders]
if UI_SETTING.enable_cache:
get_folders_sorted = st.cache_data(get_folders_sorted)
# 设置主日志路径
main_log_path = Path(args.log_dir) if args.log_dir else Path("./log")
if not main_log_path.exists():
st.error(f"Log dir {main_log_path} does not exist!")
st.stop()
if "data" not in session_state:
session_state.data = []
if "log_path" not in session_state:
session_state.log_path = None
tlist = []
def load_data():
"""加载数据到 session_state 并显示进度"""
log_file = main_log_path / session_state.log_path / "debug_llm.pkl"
try:
with st.spinner(f"正在加载数据文件 {log_file}..."):
start_time = time.time()
with open(log_file, "rb") as f:
session_state.data = pickle.load(f)
st.success(f"数据加载完成！耗时 {time.time() - start_time:.2f} 秒")
st.session_state["current_loop"] = 1
except Exception as e:
session_state.data = [{"error": str(e)}]
st.error(f"加载数据失败: {e}")
# UI - Sidebar
with st.sidebar:
st.markdown(":blue[**Log Path**]")
manually = st.toggle("Manual Input")
if manually:
st.text_input("log path", key="log_path", label_visibility="collapsed")
else:
folders = get_folders_sorted(main_log_path)
st.selectbox(f"**Select from {main_log_path.absolute()}**", folders, key="log_path")
if st.button("Refresh Data"):
load_data()
st.rerun()
# Helper functions
def show_text(text, lang=None):
"""显示文本代码块"""
if lang:
st.code(text, language=lang, wrap_lines=True)
elif "\n" in text:
st.code(text, language="python", wrap_lines=True)
else:
st.code(text, language="html", wrap_lines=True)
def highlight_prompts_uri(uri):
"""高亮 URI 的格式"""
parts = uri.split(":")
return f"**{parts[0]}:**:green[**{parts[1]}**]"
# Display Data
progress_text = st.empty()
progress_bar = st.progress(0)
# 每页展示一个 Loop
LOOPS_PER_PAGE = 1
# 获取所有的 Loop ID
loop_groups = {}
for i, d in enumerate(session_state.data):
tag = d["tag"]
loop_id, _ = extract_loopid_func_name(tag)
if loop_id:
if loop_id not in loop_groups:
loop_groups[loop_id] = []
loop_groups[loop_id].append(d)
# 按 Loop ID 排序
sorted_loop_ids = sorted(loop_groups.keys(), key=int) # 假设 Loop ID 是数字
total_loops = len(sorted_loop_ids)
total_pages = total_loops # 每页展示一个 Loop
# simple display
# FIXME: Delete this simple UI if trace have tag(evo_id & loop_id)
# with st.sidebar:
# start = int(st.text_input("start", 0))
# end = int(st.text_input("end", 100))
# for m in session_state.data[start:end]:
# if "tpl" in m["tag"]:
# obj = m["obj"]
# uri = obj["uri"]
# tpl = obj["template"]
# cxt = obj["context"]
# rd = obj["rendered"]
# with st.expander(highlight_prompts_uri(uri), expanded=False, icon="⚙️"):
# t1, t2, t3 = st.tabs([":green[**Rendered**]", ":blue[**Template**]", ":orange[**Context**]"])
# with t1:
# show_text(rd)
# with t2:
# show_text(tpl, lang="django")
# with t3:
# st.json(cxt)
# if "llm" in m["tag"]:
# obj = m["obj"]
# system = obj.get("system", None)
# user = obj["user"]
# resp = obj["resp"]
# with st.expander(f"**LLM**", expanded=False, icon="🤖"):
# t1, t2, t3 = st.tabs([":green[**Response**]", ":blue[**User**]", ":orange[**System**]"])
# with t1:
# try:
# rdict = json.loads(resp)
# if "code" in rdict:
# code = rdict["code"]
# st.markdown(":red[**Code in response dict:**]")
# st.code(code, language="python", wrap_lines=True, line_numbers=True)
# rdict.pop("code")
# elif "spec" in rdict:
# spec = rdict["spec"]
# st.markdown(":red[**Spec in response dict:**]")
# st.markdown(spec)
# rdict.pop("spec")
# else:
# # show model codes
# showed_keys = []
# for k, v in rdict.items():
# if k.startswith("model_") and k.endswith(".py"):
# st.markdown(f":red[**{k}**]")
# st.code(v, language="python", wrap_lines=True, line_numbers=True)
# showed_keys.append(k)
# for k in showed_keys:
# rdict.pop(k)
# st.write(":red[**Other parts (except for the code or spec) in response dict:**]")
# st.json(rdict)
# except:
# st.json(resp)
# with t2:
# show_text(user)
# with t3:
# show_text(system or "No system prompt available")
if total_pages:
# 初始化 current_loop
if "current_loop" not in st.session_state:
st.session_state["current_loop"] = 1
# Loop 导航按钮
col1, col2, col3, col4, col5 = st.sidebar.columns([1.2, 1, 2, 1, 1.2])
with col1:
if st.button("|<"): # 首页
st.session_state["current_loop"] = 1
with col2:
if st.button("<") and st.session_state["current_loop"] > 1: # 上一页
st.session_state["current_loop"] -= 1
with col3:
# 下拉列表显示所有 Loop
st.session_state["current_loop"] = st.selectbox(
"选择 Loop",
options=list(range(1, total_loops + 1)),
index=st.session_state["current_loop"] - 1, # 默认选中当前 Loop
label_visibility="collapsed", # 隐藏标签
)
with col4:
if st.button("\>") and st.session_state["current_loop"] < total_loops: # 下一页
st.session_state["current_loop"] += 1
with col5:
if st.button("\>|"): # 最后一页
st.session_state["current_loop"] = total_loops
# 获取当前 Loop
current_loop = st.session_state["current_loop"]
# 渲染当前 Loop 数据
loop_id = sorted_loop_ids[current_loop - 1]
progress_text = st.empty()
progress_text.text(f"正在处理 Loop {loop_id}...")
progress_bar.progress(current_loop / total_loops, text=f"Loop :green[**{current_loop}**] / {total_loops}")
# 渲染 Loop Header
loop_anchor = f"Loop_{loop_id}"
if loop_anchor not in tlist:
tlist.append(loop_anchor)
st.header(loop_anchor, anchor=loop_anchor, divider="blue")
# 渲染当前 Loop 的所有数据
loop_data = loop_groups[loop_id]
for d in loop_data:
tag = d["tag"]
obj = d["obj"]
_, func_name = extract_loopid_func_name(tag)
evo_id = extract_evoid(tag)
func_anchor = f"loop_{loop_id}.{func_name}"
if func_anchor not in tlist:
tlist.append(func_anchor)
st.header(f"in *{func_name}*", anchor=func_anchor, divider="green")
evo_anchor = f"loop_{loop_id}.evo_step_{evo_id}"
if evo_id and evo_anchor not in tlist:
tlist.append(evo_anchor)
st.subheader(f"evo_step_{evo_id}", anchor=evo_anchor, divider="orange")
# 根据 tag 渲染内容
if "debug_exp_gen" in tag:
with st.expander(
f"Exp in :violet[**{obj.experiment_workspace.workspace_path}**]", expanded=False, icon="🧩"
):
st.write(obj)
elif "debug_tpl" in tag:
uri = obj["uri"]
tpl = obj["template"]
cxt = obj["context"]
rd = obj["rendered"]
with st.expander(highlight_prompts_uri(uri), expanded=False, icon="⚙️"):
t1, t2, t3 = st.tabs([":green[**Rendered**]", ":blue[**Template**]", ":orange[**Context**]"])
with t1:
show_text(rd)
with t2:
show_text(tpl, lang="django")
with t3:
st.json(cxt)
elif "debug_llm" in tag:
system = obj.get("system", None)
user = obj["user"]
resp = obj["resp"]
with st.expander(f"**LLM**", expanded=False, icon="🤖"):
t1, t2, t3 = st.tabs([":green[**Response**]", ":blue[**User**]", ":orange[**System**]"])
with t1:
try:
rdict = json.loads(resp)
if "code" in rdict:
code = rdict["code"]
st.markdown(":red[**Code in response dict:**]")
st.code(code, language="python", wrap_lines=True, line_numbers=True)
rdict.pop("code")
elif "spec" in rdict:
spec = rdict["spec"]
st.markdown(":red[**Spec in response dict:**]")
st.markdown(spec)
rdict.pop("spec")
else:
# show model codes
showed_keys = []
for k, v in rdict.items():
if k.startswith("model_") and k.endswith(".py"):
st.markdown(f":red[**{k}**]")
st.code(v, language="python", wrap_lines=True, line_numbers=True)
showed_keys.append(k)
for k in showed_keys:
rdict.pop(k)
st.write(":red[**Other parts (except for the code or spec) in response dict:**]")
st.json(rdict)
except:
st.json(resp)
with t2:
show_text(user)
with t3:
show_text(system or "No system prompt available")
progress_text.text("当前 Loop 数据处理完成！")
# Sidebar TOC
with st.sidebar:
toc = "\n".join([f"- [{t}](#{t})" if t.startswith("L") else f" - [{t.split('.')[1]}](#{t})" for t in tlist])
st.markdown(toc, unsafe_allow_html=True)
rdagent/log/ui/qlib_report_figure.py
-445
View File
@@ -1,445 +0,0 @@
import importlib
import math
import pandas as pd
import plotly.graph_objs as go
from plotly.subplots import make_subplots
class BaseGraph:
_name = None
def __init__(
self, df: pd.DataFrame = None, layout: dict = None, graph_kwargs: dict = None, name_dict: dict = None, **kwargs
):
"""
:param df:
:param layout:
:param graph_kwargs:
:param name_dict:
:param kwargs:
layout: dict
go.Layout parameters
graph_kwargs: dict
Graph parameters, eg: go.Bar(**graph_kwargs)
"""
self._df = df
self._layout = dict() if layout is None else layout
self._graph_kwargs = dict() if graph_kwargs is None else graph_kwargs
self._name_dict = name_dict
self.data = None
self._init_parameters(**kwargs)
self._init_data()
def _init_data(self):
"""
:return:
"""
if self._df.empty:
raise ValueError("df is empty.")
self.data = self._get_data()
def _init_parameters(self, **kwargs):
"""
:param kwargs
"""
# Instantiate graphics parameters
self._graph_type = self._name.lower().capitalize()
# Displayed column name
if self._name_dict is None:
self._name_dict = {_item: _item for _item in self._df.columns}
@staticmethod
def get_instance_with_graph_parameters(graph_type: str = None, **kwargs):
"""
:param graph_type:
:param kwargs:
:return:
"""
try:
_graph_module = importlib.import_module("plotly.graph_objs")
_graph_class = getattr(_graph_module, graph_type)
except AttributeError:
_graph_module = importlib.import_module("qlib.contrib.report.graph")
_graph_class = getattr(_graph_module, graph_type)
return _graph_class(**kwargs)
def _get_layout(self) -> go.Layout:
"""
:return:
"""
return go.Layout(**self._layout)
def _get_data(self) -> list:
"""
:return:
"""
_data = [
self.get_instance_with_graph_parameters(
graph_type=self._graph_type, x=self._df.index, y=self._df[_col], name=_name, **self._graph_kwargs
)
for _col, _name in self._name_dict.items()
]
return _data
@property
def figure(self) -> go.Figure:
"""
:return:
"""
_figure = go.Figure(data=self.data, layout=self._get_layout())
# NOTE: Use the default theme from plotly version 3.x, template=None
_figure["layout"].update(template=None)
return _figure
class SubplotsGraph:
"""Create subplots same as df.plot(subplots=True)
Simple package for `plotly.tools.subplots`
"""
def __init__(
self,
df: pd.DataFrame = None,
kind_map: dict = None,
layout: dict = None,
sub_graph_layout: dict = None,
sub_graph_data: list = None,
subplots_kwargs: dict = None,
**kwargs,
):
"""
:param df: pd.DataFrame
:param kind_map: dict, subplots graph kind and kwargs
eg: dict(kind='Scatter', kwargs=dict())
:param layout: `go.Layout` parameters
:param sub_graph_layout: Layout of each graphic, similar to 'layout'
:param sub_graph_data: Instantiation parameters for each sub-graphic
eg: [(column_name, instance_parameters), ]
column_name: str or go.Figure
Instance_parameters:
- row: int, the row where the graph is located
- col: int, the col where the graph is located
- name: str, show name, default column_name in 'df'
- kind: str, graph kind, default `kind` param, eg: bar, scatter, ...
- graph_kwargs: dict, graph kwargs, default {}, used in `go.Bar(**graph_kwargs)`
:param subplots_kwargs: `plotly.tools.make_subplots` original parameters
- shared_xaxes: bool, default False
- shared_yaxes: bool, default False
- vertical_spacing: float, default 0.3 / rows
- subplot_titles: list, default []
If `sub_graph_data` is None, will generate 'subplot_titles' according to `df.columns`,
this field will be discarded
- specs: list, see `make_subplots` docs
- rows: int, Number of rows in the subplot grid, default 1
If `sub_graph_data` is None, will generate 'rows' according to `df`, this field will be discarded
- cols: int, Number of cols in the subplot grid, default 1
If `sub_graph_data` is None, will generate 'cols' according to `df`, this field will be discarded
:param kwargs:
"""
self._df = df
self._layout = layout
self._sub_graph_layout = sub_graph_layout
self._kind_map = kind_map
if self._kind_map is None:
self._kind_map = dict(kind="Scatter", kwargs=dict())
self._subplots_kwargs = subplots_kwargs
if self._subplots_kwargs is None:
self._init_subplots_kwargs()
self.__cols = self._subplots_kwargs.get("cols", 2) # pylint: disable=W0238
self.__rows = self._subplots_kwargs.get( # pylint: disable=W0238
"rows", math.ceil(len(self._df.columns) / self.__cols)
)
self._sub_graph_data = sub_graph_data
if self._sub_graph_data is None:
self._init_sub_graph_data()
self._init_figure()
def _init_sub_graph_data(self):
"""
:return:
"""
self._sub_graph_data = []
self._subplot_titles = []
for i, column_name in enumerate(self._df.columns):
row = math.ceil((i + 1) / self.__cols)
_temp = (i + 1) % self.__cols
col = _temp if _temp else self.__cols
res_name = column_name.replace("_", " ")
_temp_row_data = (
column_name,
dict(
row=row,
col=col,
name=res_name,
kind=self._kind_map["kind"],
graph_kwargs=self._kind_map["kwargs"],
),
)
self._sub_graph_data.append(_temp_row_data)
self._subplot_titles.append(res_name)
def _init_subplots_kwargs(self):
"""
:return:
"""
# Default cols, rows
_cols = 2
_rows = math.ceil(len(self._df.columns) / 2)
self._subplots_kwargs = dict()
self._subplots_kwargs["rows"] = _rows
self._subplots_kwargs["cols"] = _cols
self._subplots_kwargs["shared_xaxes"] = False
self._subplots_kwargs["shared_yaxes"] = False
self._subplots_kwargs["vertical_spacing"] = 0.3 / _rows
self._subplots_kwargs["print_grid"] = False
self._subplots_kwargs["subplot_titles"] = self._df.columns.tolist()
def _init_figure(self):
"""
:return:
"""
self._figure = make_subplots(**self._subplots_kwargs)
for column_name, column_map in self._sub_graph_data:
if isinstance(column_name, go.Figure):
_graph_obj = column_name
elif isinstance(column_name, str):
temp_name = column_map.get("name", column_name.replace("_", " "))
kind = column_map.get("kind", self._kind_map.get("kind", "Scatter"))
_graph_kwargs = column_map.get("graph_kwargs", self._kind_map.get("kwargs", {}))
_graph_obj = BaseGraph.get_instance_with_graph_parameters(
kind,
**dict(
x=self._df.index,
y=self._df[column_name],
name=temp_name,
**_graph_kwargs,
),
)
else:
raise TypeError()
row = column_map["row"]
col = column_map["col"]
self._figure.add_trace(_graph_obj, row=row, col=col)
if self._sub_graph_layout is not None:
for k, v in self._sub_graph_layout.items():
self._figure["layout"][k].update(v)
# NOTE: Use the default theme from plotly version 3.x: template=None
self._figure["layout"].update(template=None)
self._figure["layout"].update(self._layout)
@property
def figure(self):
return self._figure
def _calculate_maximum(df: pd.DataFrame, is_ex: bool = False):
"""
:param df:
:param is_ex:
:return:
"""
if is_ex:
end_date = df["cum_ex_return_wo_cost_mdd"].idxmin()
start_date = df.loc[df.index <= end_date]["cum_ex_return_wo_cost"].idxmax()
else:
end_date = df["return_wo_mdd"].idxmin()
start_date = df.loc[df.index <= end_date]["cum_return_wo_cost"].idxmax()
return start_date, end_date
def _calculate_mdd(series):
"""
Calculate mdd
:param series:
:return:
"""
return series - series.cummax()
def _calculate_report_data(raw_df: pd.DataFrame) -> pd.DataFrame:
"""
:param df:
:return:
"""
df = raw_df.copy(deep=True)
index_names = df.index.names
df.index = df.index.strftime("%Y-%m-%d")
report_df = pd.DataFrame()
report_df["cum_bench"] = df["bench"].cumsum()
report_df["cum_return_wo_cost"] = df["return"].cumsum()
report_df["cum_return_w_cost"] = (df["return"] - df["cost"]).cumsum()
# report_df['cum_return'] - report_df['cum_return'].cummax()
report_df["return_wo_mdd"] = _calculate_mdd(report_df["cum_return_wo_cost"])
report_df["return_w_cost_mdd"] = _calculate_mdd((df["return"] - df["cost"]).cumsum())
report_df["cum_ex_return_wo_cost"] = (df["return"] - df["bench"]).cumsum()
report_df["cum_ex_return_w_cost"] = (df["return"] - df["bench"] - df["cost"]).cumsum()
report_df["cum_ex_return_wo_cost_mdd"] = _calculate_mdd((df["return"] - df["bench"]).cumsum())
report_df["cum_ex_return_w_cost_mdd"] = _calculate_mdd((df["return"] - df["cost"] - df["bench"]).cumsum())
# return_wo_mdd , return_w_cost_mdd, cum_ex_return_wo_cost_mdd, cum_ex_return_w
report_df["turnover"] = df["turnover"]
report_df.sort_index(ascending=True, inplace=True)
report_df.index.names = index_names
return report_df
def report_figure(df: pd.DataFrame) -> list | tuple:
"""
:param df:
:return:
"""
# Get data
report_df = _calculate_report_data(df)
# Maximum Drawdown
max_start_date, max_end_date = _calculate_maximum(report_df)
ex_max_start_date, ex_max_end_date = _calculate_maximum(report_df, True)
index_name = report_df.index.name
_temp_df = report_df.reset_index()
_temp_df.loc[-1] = 0
_temp_df = _temp_df.shift(1)
_temp_df.loc[0, index_name] = "T0"
_temp_df.set_index(index_name, inplace=True)
_temp_df.iloc[0] = 0
report_df = _temp_df
# Create figure
_default_kind_map = dict(kind="Scatter", kwargs={"mode": "lines+markers"})
_temp_fill_args = {"fill": "tozeroy", "mode": "lines+markers"}
_column_row_col_dict = [
("cum_bench", dict(row=1, col=1)),
("cum_return_wo_cost", dict(row=1, col=1)),
("cum_return_w_cost", dict(row=1, col=1)),
("return_wo_mdd", dict(row=2, col=1, graph_kwargs=_temp_fill_args)),
("return_w_cost_mdd", dict(row=3, col=1, graph_kwargs=_temp_fill_args)),
("cum_ex_return_wo_cost", dict(row=4, col=1)),
("cum_ex_return_w_cost", dict(row=4, col=1)),
("turnover", dict(row=5, col=1)),
("cum_ex_return_w_cost_mdd", dict(row=6, col=1, graph_kwargs=_temp_fill_args)),
("cum_ex_return_wo_cost_mdd", dict(row=7, col=1, graph_kwargs=_temp_fill_args)),
]
_subplot_layout = dict()
for i in range(1, 8):
# yaxis
_subplot_layout.update({"yaxis{}".format(i): dict(zeroline=True, showline=True, showticklabels=True)})
_show_line = i == 7
_subplot_layout.update({"xaxis{}".format(i): dict(showline=_show_line, type="category", tickangle=45)})
_layout_style = dict(
height=1200,
title=" ",
shapes=[
{
"type": "rect",
"xref": "x",
"yref": "paper",
"x0": max_start_date,
"y0": 0.55,
"x1": max_end_date,
"y1": 1,
"fillcolor": "#d3d3d3",
"opacity": 0.3,
"line": {
"width": 0,
},
},
{
"type": "rect",
"xref": "x",
"yref": "paper",
"x0": ex_max_start_date,
"y0": 0,
"x1": ex_max_end_date,
"y1": 0.55,
"fillcolor": "#d3d3d3",
"opacity": 0.3,
"line": {
"width": 0,
},
},
],
)
_subplot_kwargs = dict(
shared_xaxes=True,
vertical_spacing=0.01,
rows=7,
cols=1,
row_width=[1, 1, 1, 3, 1, 1, 3],
print_grid=False,
)
figure = SubplotsGraph(
df=report_df,
layout=_layout_style,
sub_graph_data=_column_row_col_dict,
subplots_kwargs=_subplot_kwargs,
kind_map=_default_kind_map,
sub_graph_layout=_subplot_layout,
).figure
return figure
rdagent/log/ui/st_fixed_container.py
-126
View File
@@ -1,126 +0,0 @@
from typing import Literal
import streamlit as st
from streamlit.components.v1 import html
FIXED_CONTAINER_CSS = """
:root {{
--background-color: #ffffff; /* Default background color */
}}
div[data-testid="stVerticalBlockBorderWrapper"]:has(div.fixed-container-{id}):not(:has(div.not-fixed-container)) {{
position: {mode};
width: inherit;
background-color: inherit;
{position}: {margin};
z-index: 999;
}}
div[data-testid="stVerticalBlockBorderWrapper"]:has(div.fixed-container-{id}):not(:has(div.not-fixed-container)) div[data-testid="stVerticalBlock"]:has(div.fixed-container-{id}):not(:has(div.not-fixed-container)) > div[data-testid="stVerticalBlockBorderWrapper"] {{
background-color: transparent;
width: 100%;
}}
div[data-testid="stVerticalBlockBorderWrapper"]:has(div.fixed-container-{id}):not(:has(div.not-fixed-container)) div[data-testid="stVerticalBlock"]:has(div.fixed-container-{id}):not(:has(div.not-fixed-container)) > div[data-testid="stVerticalBlockBorderWrapper"] div[data-testid="stVerticalBlockBorderWrapper"] {{
background-color: var(--background-color);
}}
div[data-testid="stVerticalBlockBorderWrapper"]:has(div.fixed-container-{id}):not(:has(div.not-fixed-container)) div[data-testid="stVerticalBlock"]:has(div.fixed-container-{id}):not(:has(div.not-fixed-container)) > div[data-testid="element-container"] {{
display: none;
}}
div[data-testid="stVerticalBlockBorderWrapper"]:has(div.not-fixed-container):not(:has(div[class^='fixed-container-'])) {{
display: none;
}}
""".strip()
FIXED_CONTAINER_JS = """
const root = parent.document.querySelector('.stApp');
let lastBackgroundColor = null;
function updateContainerBackground(currentBackground) {
parent.document.documentElement.style.setProperty('--background-color', currentBackground);
;
}
function checkForBackgroundColorChange() {
const style = window.getComputedStyle(root);
const currentBackgroundColor = style.backgroundColor;
if (currentBackgroundColor !== lastBackgroundColor) {
lastBackgroundColor = currentBackgroundColor; // Update the last known value
updateContainerBackground(lastBackgroundColor);
}
}
const observerCallback = (mutationsList, observer) => {
for(let mutation of mutationsList) {
if (mutation.type === 'attributes' && (mutation.attributeName === 'class' || mutation.attributeName === 'style')) {
checkForBackgroundColorChange();
}
}
};
const main = () => {
checkForBackgroundColorChange();
const observer = new MutationObserver(observerCallback);
observer.observe(root, { attributes: true, childList: false, subtree: false });
}
// main();
document.addEventListener("DOMContentLoaded", main);
""".strip()
MARGINS = {
"top": "2.875rem",
"bottom": "0",
}
counter = 0
def st_fixed_container(
*,
height: int | None = None,
border: bool | None = None,
mode: Literal["fixed", "sticky"] = "fixed",
position: Literal["top", "bottom"] = "top",
margin: str | None = None,
transparent: bool = False,
):
if margin is None:
margin = MARGINS[position]
global counter
fixed_container = st.container()
non_fixed_container = st.container()
css = FIXED_CONTAINER_CSS.format(
mode=mode,
position=position,
margin=margin,
id=counter,
)
with fixed_container:
html(f"<script>{FIXED_CONTAINER_JS}</script>", scrolling=False, height=0)
st.markdown(f"<style>{css}</style>", unsafe_allow_html=True)
st.markdown(
f"<div class='fixed-container-{counter}'></div>",
unsafe_allow_html=True,
)
with non_fixed_container:
st.markdown(
f"<div class='not-fixed-container'></div>",
unsafe_allow_html=True,
)
counter += 1
parent_container = fixed_container if transparent else fixed_container.container()
return parent_container.container(height=height, border=border)
if __name__ == "__main__":
for i in range(30):
st.write(f"Line {i}")
# with st_fixed_container(mode="sticky", position="top", border=True):
# with st_fixed_container(mode="sticky", position="bottom", border=True):
# with st_fixed_container(mode="fixed", position="top", border=True):
with st_fixed_container(mode="fixed", position="bottom", border=True):
st.write("This is a fixed container.")
st.write("This is a fixed container.")
st.write("This is a fixed container.")
st.container(border=True).write("This is a regular container.")
for i in range(30):
st.write(f"Line {i}")
rdagent/log/ui/storage.py
-345
View File
@@ -1,345 +0,0 @@
from datetime import datetime
from pathlib import Path
from typing import Any, Generator
import requests
from rdagent.log.base import Message, Storage
from rdagent.log.utils import extract_evoid, extract_loopid_func_name, gen_datetime
from .conf import UI_SETTING
class WebStorage(Storage):
"""
The storage for web app.
It is used to provide the data for the web app.
"""
def __init__(self, port: int, path: str) -> None:
"""
Initializes the storage object with the specified port and identifier.
Args:
port (int): The port number to use for the storage service.
path (str): The unique identifier for local storage, the log path.
"""
self.url = f"http://localhost:{port}"
self.path = path
self.msgs = []
def __str__(self):
return f"WebStorage({self.url})"
def log(self, obj: object, tag: str, timestamp: datetime | None = None, **kwargs: Any) -> str | Path:
timestamp = gen_datetime(timestamp)
if "pdf_image" in tag or "load_pdf_screenshot" in tag:
Path(f"{UI_SETTING.static_path}/pdf_images").mkdir(parents=True, exist_ok=True)
obj.save(f"{UI_SETTING.static_path}/pdf_images/{timestamp.isoformat()}.jpg")
try:
data = self._obj_to_json(obj=obj, tag=tag, id=str(self.path), timestamp=timestamp.isoformat())
if not data:
return "Normal log, skipped"
if isinstance(data, list):
for d in data:
self.msgs.append(d)
else:
self.msgs.append(data)
headers = {"Content-Type": "application/json"}
resp = requests.post(f"{self.url}/receive", json=data, headers=headers, timeout=1)
return f"{resp.status_code} {resp.text}"
except (requests.ConnectionError, requests.Timeout) as e:
print(f"Failed to connect to the web storage server at {self.url}: {e}")
def truncate(self, time: datetime) -> None:
self.msgs = [m for m in self.msgs if datetime.fromisoformat(m["msg"]["timestamp"]) <= time]
def iter_msg(self, **kwargs: Any) -> Generator[Message, None, None]:
for msg in self.msgs:
yield Message(
tag=msg["msg"]["tag"],
level="INFO",
timestamp=datetime.fromisoformat(msg["msg"]["timestamp"]),
content=msg,
)
def _obj_to_json(
self,
obj: object,
tag: str,
id: str,
timestamp: str,
) -> list[dict] | dict:
li, fn = extract_loopid_func_name(tag)
ei = extract_evoid(tag)
data = {}
if "hypothesis generation" in tag:
from rdagent.core.proposal import Hypothesis
h: Hypothesis = obj
data = {
"id": id,
"msg": {
"tag": "research.hypothesis",
"timestamp": timestamp,
"loop_id": li,
"content": {
"hypothesis": h.hypothesis,
"reason": h.reason,
"concise_reason": h.concise_reason,
"concise_justification": h.concise_justification,
"concise_observation": h.concise_observation,
"concise_knowledge": h.concise_knowledge,
},
},
}
elif "pdf_image" in tag or "load_pdf_screenshot" in tag:
# obj.save(f"{app.static_folder}/{timestamp}.jpg")
data = {
"id": id,
"msg": {
"tag": "research.pdf_image",
"timestamp": timestamp,
"loop_id": li,
"content": {"image": f"pdf_images/{timestamp}.jpg"},
},
}
elif "experiment generation" in tag or "load_experiment" in tag:
from rdagent.components.coder.factor_coder.factor import FactorTask
from rdagent.components.coder.model_coder.model import ModelTask
if "load_experiment" in tag:
tasks: list[FactorTask | ModelTask] = obj.sub_tasks
else:
tasks: list[FactorTask | ModelTask] = obj
if isinstance(tasks[0], FactorTask):
data = {
"id": id,
"msg": {
"tag": "research.tasks",
"timestamp": timestamp,
"loop_id": li,
"content": [
{
"name": t.factor_name,
"description": t.factor_description,
"formulation": t.factor_formulation,
"variables": t.variables,
}
for t in tasks
],
},
}
elif isinstance(tasks[0], ModelTask):
data = {
"id": id,
"msg": {
"tag": "research.tasks",
"timestamp": timestamp,
"loop_id": li,
"content": [
{
"name": t.name,
"description": t.description,
"model_type": t.model_type,
"formulation": t.formulation,
"variables": t.variables,
}
for t in tasks
],
},
}
elif "direct_exp_gen" in tag:
from rdagent.scenarios.data_science.experiment.experiment import (
DSExperiment,
)
if isinstance(obj, DSExperiment):
from rdagent.scenarios.data_science.proposal.exp_gen.base import (
DSHypothesis,
)
h: DSHypothesis = obj.hypothesis
tasks = [t[0] for t in obj.pending_tasks_list]
t = tasks[0]
t.name = type(t).__name__ # TODO: PipelinTask have "COMPONENT" in name, fix this when creating task.
data = [
{
"id": id,
"msg": {
"tag": "research.hypothesis",
"old_tag": tag,
"timestamp": timestamp,
"loop_id": li,
"content": {
"name_map": {
"hypothesis": "RD-Agent proposes the hypothesis⬇️",
"concise_justification": "because the reason⬇️",
"concise_observation": "based on the observation⬇️",
"concise_knowledge": "Knowledge⬇️ gained after practice",
"no_hypothesis": f"No hypothesis available. Trying to construct the first runnable {h.component} component.",
},
"hypothesis": h.hypothesis,
"reason": h.reason,
"component": h.component,
"concise_reason": h.concise_reason,
"concise_justification": h.concise_justification,
"concise_observation": h.concise_observation,
"concise_knowledge": h.concise_knowledge,
},
},
},
{
"id": id,
"msg": {
"tag": "research.tasks",
"old_tag": tag,
"timestamp": timestamp,
"loop_id": li,
"content": [
(
{
"name": t.name,
"description": t.description,
}
if not hasattr(t, "architecture")
else {
"name": t.name,
"description": t.description,
"model_type": t.model_type,
"architecture": t.architecture,
"hyperparameters": t.hyperparameters,
}
)
],
},
},
]
elif f"evo_loop_{ei}.evolving code" in tag and "running" not in tag:
from rdagent.core.experiment import FBWorkspace
ws: list[FBWorkspace] = [i for i in obj]
data = {
"id": id,
"msg": {
"tag": "evolving.codes",
"timestamp": timestamp,
"loop_id": li,
"evo_id": ei,
"content": [
{
"evo_id": ei,
"target_task_name": (
w.target_task.name if w.target_task else "PipelineTask"
), # TODO: save this when proposal
"workspace": w.file_dict,
}
for w in ws
],
},
}
elif f"evo_loop_{ei}.evolving feedback" in tag and "running" not in tag:
from rdagent.components.coder.CoSTEER.evaluators import (
CoSTEERSingleFeedback,
)
fl: list[CoSTEERSingleFeedback] = [i for i in obj]
data = {
"id": id,
"msg": {
"tag": "evolving.feedbacks",
"timestamp": timestamp,
"loop_id": li,
"evo_id": ei,
"content": [
{
"evo_id": ei,
"final_decision": f.final_decision,
# "final_feedback": f.final_feedback,
"execution": f.execution,
"code": f.code,
"return_checking": f.return_checking,
}
for f in fl
],
},
}
elif "scenario" in tag:
data = {
"id": id,
"msg": {
"tag": "feedback.config",
"timestamp": timestamp,
"loop_id": li,
"content": {"config": obj.experiment_setting},
},
}
elif "Quantitative Backtesting Chart" in tag:
import plotly
from rdagent.log.ui.qlib_report_figure import report_figure
data = {
"id": id,
"msg": {
"tag": "feedback.return_chart",
"timestamp": timestamp,
"loop_id": li,
"content": {"chart_html": plotly.io.to_html(report_figure(obj))},
},
}
elif "running" in tag:
from rdagent.core.experiment import Experiment
if isinstance(obj, Experiment):
try:
result = obj.result
except AttributeError: # compatibility with old versions
result = obj.__dict__["result"]
if result is not None:
result_str = result.to_json()
data = {
"id": id,
"msg": {
"tag": "feedback.metric",
"old_tag": tag,
"timestamp": timestamp,
"loop_id": li,
"content": {
"result": result_str,
},
},
}
elif "feedback" in tag:
from rdagent.core.proposal import ExperimentFeedback, HypothesisFeedback
if isinstance(obj, ExperimentFeedback):
ef: ExperimentFeedback = obj
content = (
{
"observations": str(ef.observations),
"hypothesis_evaluation": ef.hypothesis_evaluation,
"new_hypothesis": ef.new_hypothesis,
"decision": ef.decision,
"reason": ef.reason,
"exception": ef.exception,
}
if isinstance(ef, HypothesisFeedback)
else {
"decision": ef.decision,
"reason": ef.reason,
"exception": ef.exception,
}
)
data = {
"id": id,
"msg": {
"tag": "feedback.hypothesis_feedback",
"timestamp": timestamp,
"loop_id": li,
"content": content,
},
}
return data
rdagent/log/ui/utils.py
-1128
View File
File diff suppressed because it is too large Load Diff
rdagent/log/ui/web.py
-629
View File
@@ -1,629 +0,0 @@
import time
from collections import defaultdict
from copy import deepcopy
from datetime import datetime, timezone
from typing import Callable, Type
import pandas as pd
import plotly.express as px
import streamlit as st
from streamlit.delta_generator import DeltaGenerator
from rdagent.components.coder.factor_coder.evaluators import FactorSingleFeedback
from rdagent.components.coder.factor_coder.factor import FactorFBWorkspace, FactorTask
from rdagent.components.coder.model_coder.evaluators import ModelSingleFeedback
from rdagent.components.coder.model_coder.model import ModelFBWorkspace, ModelTask
from rdagent.core.proposal import Hypothesis, HypothesisFeedback, Trace
from rdagent.log.base import Message, Storage, View
from rdagent.scenarios.qlib.experiment.factor_experiment import QlibFactorExperiment
from rdagent.scenarios.qlib.experiment.model_experiment import (
QlibModelExperiment,
QlibModelScenario,
)
st.set_page_config(layout="wide")
TIME_DELAY = 0.001
class WebView(View):
def __init__(self, ui: "StWindow"):
self.ui = ui
# Save logs to your desired data structure
# ...
def display(self, s: Storage, watch: bool = False):
for msg in s.iter_msg(): # iterate overtime
# NOTE: iter_msg will correctly separate the information.
# TODO: msg may support streaming mode.
self.ui.consume_msg(msg)
class StWindow:
def __init__(self, container: "DeltaGenerator"):
self.container = container
def consume_msg(self, msg: Message):
msg_str = f"{msg.timestamp.astimezone(timezone.utc).isoformat()} | {msg.level} | {msg.caller} - {msg.content}"
self.container.code(msg_str, language="log")
class LLMWindow(StWindow):
def __init__(self, container: "DeltaGenerator", session_name: str = "common"):
self.session_name = session_name
self.container = container.expander(f"{self.session_name} message")
def consume_msg(self, msg: Message):
self.container.chat_message("user").markdown(f"{msg.content}")
class ProgressTabsWindow(StWindow):
"""
For windows with stream messages, will refresh when a new tab is created.
"""
def __init__(
self,
container: "DeltaGenerator",
inner_class: Type[StWindow] = StWindow,
mapper: Callable[[Message], str] = lambda x: x.pid_trace,
):
self.inner_class = inner_class
self.mapper = mapper
self.container = container.empty()
self.tab_windows: dict[str, StWindow] = defaultdict(None)
self.tab_caches: dict[str, list[Message]] = defaultdict(list)
def consume_msg(self, msg: Message):
name = self.mapper(msg)
if name not in self.tab_windows:
# new tab need to be created, current streamlit container need to be updated.
names = list(self.tab_windows.keys()) + [name]
if len(names) == 1:
tabs = [self.container.container()]
else:
tabs = self.container.tabs(names)
for id, name in enumerate(names):
self.tab_windows[name] = self.inner_class(tabs[id])
# consume the cache
for name in self.tab_caches:
for msg in self.tab_caches[name]:
self.tab_windows[name].consume_msg(msg)
self.tab_caches[name].append(msg)
self.tab_windows[name].consume_msg(msg)
class ObjectsTabsWindow(StWindow):
def __init__(
self,
container: "DeltaGenerator",
inner_class: Type[StWindow] = StWindow,
mapper: Callable[[object], str] = lambda x: str(x),
tab_names: list[str] | None = None,
):
self.inner_class = inner_class
self.mapper = mapper
self.container = container
self.tab_names = tab_names
def consume_msg(self, msg: Message):
if isinstance(msg.content, list):
if self.tab_names:
assert len(self.tab_names) == len(
msg.content
), "List of objects should have the same length as provided tab names."
objs_dict = {self.tab_names[id]: obj for id, obj in enumerate(msg.content)}
else:
objs_dict = {self.mapper(obj): obj for obj in msg.content}
elif not isinstance(msg.content, dict):
raise ValueError("Message content should be a list or a dict of objects.")
# two many tabs may cause display problem
tab_names = list(objs_dict.keys())
tabs = []
for i in range(0, len(tab_names), 10):
tabs.extend(self.container.tabs(tab_names[i : i + 10]))
for id, obj in enumerate(objs_dict.values()):
splited_msg = Message(
tag=msg.tag,
level=msg.level,
timestamp=msg.timestamp,
caller=msg.caller,
pid_trace=msg.pid_trace,
content=obj,
)
self.inner_class(tabs[id]).consume_msg(splited_msg)
class RoundTabsWindow(StWindow):
def __init__(
self,
container: "DeltaGenerator",
new_tab_func: Callable[[Message], bool],
inner_class: Type[StWindow] = StWindow,
title: str = "Round tabs",
):
container.markdown(f"### **{title}**")
self.inner_class = inner_class
self.new_tab_func = new_tab_func
self.round = 0
self.current_win = StWindow(container)
self.tabs_c = container.empty()
def consume_msg(self, msg: Message):
if self.new_tab_func(msg):
self.round += 1
self.current_win = self.inner_class(self.tabs_c.tabs([str(i) for i in range(1, self.round + 1)])[-1])
self.current_win.consume_msg(msg)
class HypothesisWindow(StWindow):
def consume_msg(self, msg: Message | Hypothesis):
h: Hypothesis = msg.content if isinstance(msg, Message) else msg
self.container.markdown("#### **Hypothesis💡**")
self.container.markdown(f"""
- **Hypothesis**: {h.hypothesis}
- **Reason**: {h.reason}""")
class HypothesisFeedbackWindow(StWindow):
def consume_msg(self, msg: Message | HypothesisFeedback):
h: HypothesisFeedback = msg.content if isinstance(msg, Message) else msg
self.container.markdown("#### **Hypothesis Feedback🔍**")
self.container.markdown(f"""
- **Observations**: {h.observations}
- **Hypothesis Evaluation**: {h.hypothesis_evaluation}
- **New Hypothesis**: {h.new_hypothesis}
- **Decision**: {h.decision}
- **Reason**: {h.reason}""")
class FactorTaskWindow(StWindow):
def consume_msg(self, msg: Message | FactorTask):
ft: FactorTask = msg.content if isinstance(msg, Message) else msg
self.container.markdown(f"**Factor Name**: {ft.factor_name}")
self.container.markdown(f"**Description**: {ft.factor_description}")
self.container.latex(f"Formulation: {ft.factor_formulation}")
variables_df = pd.DataFrame(ft.variables, index=["Description"]).T
variables_df.index.name = "Variable"
self.container.table(variables_df)
self.container.text(f"Factor resources: {ft.factor_resources}")
class ModelTaskWindow(StWindow):
def consume_msg(self, msg: Message | ModelTask):
mt: ModelTask = msg.content if isinstance(msg, Message) else msg
self.container.markdown(f"**Model Name**: {mt.name}")
self.container.markdown(f"**Model Type**: {mt.model_type}")
self.container.markdown(f"**Description**: {mt.description}")
self.container.latex(f"Formulation: {mt.formulation}")
variables_df = pd.DataFrame(mt.variables, index=["Value"]).T
variables_df.index.name = "Variable"
self.container.table(variables_df)
class FactorFeedbackWindow(StWindow):
def consume_msg(self, msg: Message | FactorSingleFeedback):
fb: FactorSingleFeedback = msg.content if isinstance(msg, Message) else msg
self.container.markdown(f"""### :blue[Factor Execution Feedback]
{fb.execution_feedback}
### :blue[Factor Code Feedback]
{fb.code_feedback}
### :blue[Factor Value Feedback]
{fb.value_feedback}
### :blue[Factor Final Feedback]
{fb.final_feedback}
### :blue[Factor Final Decision]
This implementation is {'SUCCESS' if fb.final_decision else 'FAIL'}.
""")
class ModelFeedbackWindow(StWindow):
def consume_msg(self, msg: Message | ModelSingleFeedback):
mb: ModelSingleFeedback = msg.content if isinstance(msg, Message) else msg
self.container.markdown(f"""### :blue[Model Execution Feedback]
{mb.execution_feedback}
### :blue[Model Shape Feedback]
{mb.shape_feedback}
### :blue[Model Value Feedback]
{mb.value_feedback}
### :blue[Model Code Feedback]
{mb.code_feedback}
### :blue[Model Final Feedback]
{mb.final_feedback}
### :blue[Model Final Decision]
This implementation is {'SUCCESS' if mb.final_decision else 'FAIL'}.
""")
class WorkspaceWindow(StWindow):
def __init__(self, container: "DeltaGenerator", show_task_info: bool = False):
self.container = container
self.show_task_info = show_task_info
def consume_msg(self, msg: Message | FactorFBWorkspace | ModelFBWorkspace):
ws: FactorFBWorkspace | ModelFBWorkspace = msg.content if isinstance(msg, Message) else msg
# no workspace
if ws is None:
return
# task info
if self.show_task_info:
task_msg = deepcopy(msg)
task_msg.content = ws.target_task
if isinstance(ws, FactorFBWorkspace):
self.container.subheader("Factor Info")
FactorTaskWindow(self.container.container()).consume_msg(task_msg)
else:
self.container.subheader("Model Info")
ModelTaskWindow(self.container.container()).consume_msg(task_msg)
# task codes
for k, v in ws.file_dict.items():
self.container.markdown(f"`{k}`")
self.container.code(v, language="python")
class QlibFactorExpWindow(StWindow):
def __init__(self, container: DeltaGenerator, show_task_info: bool = False):
self.container = container
self.show_task_info = show_task_info
def consume_msg(self, msg: Message | QlibFactorExperiment):
exp: QlibFactorExperiment = msg.content if isinstance(msg, Message) else msg
# factor tasks
if self.show_task_info:
ftm_msg = deepcopy(msg)
ftm_msg.content = [ws for ws in exp.sub_workspace_list if ws]
self.container.markdown("**Factor Tasks**")
ObjectsTabsWindow(
self.container.container(),
inner_class=WorkspaceWindow,
mapper=lambda x: x.target_task.factor_name,
).consume_msg(ftm_msg)
# result
self.container.markdown("**Results**")
results = pd.DataFrame({f"base_exp_{id}": e.result for id, e in enumerate(exp.based_experiments)})
results["now"] = exp.result
self.container.expander("results table").table(results)
try:
bar_chart = px.bar(results, orientation="h", barmode="group")
self.container.expander("results chart").plotly_chart(bar_chart)
except:
self.container.text("Results are incomplete.")
class QlibModelExpWindow(StWindow):
def __init__(self, container: DeltaGenerator, show_task_info: bool = False):
self.container = container
self.show_task_info = show_task_info
def consume_msg(self, msg: Message | QlibModelExperiment):
exp: QlibModelExperiment = msg.content if isinstance(msg, Message) else msg
# model tasks
if self.show_task_info:
_msg = deepcopy(msg)
_msg.content = [ws for ws in exp.sub_workspace_list if ws]
self.container.markdown("**Model Tasks**")
ObjectsTabsWindow(
self.container.container(),
inner_class=WorkspaceWindow,
mapper=lambda x: x.target_task.name,
).consume_msg(_msg)
# result
self.container.subheader("Results", divider=True)
results = pd.DataFrame({f"base_exp_{id}": e.result for id, e in enumerate(exp.based_experiments)})
results["now"] = exp.result
self.container.expander("results table").table(results)
class SimpleTraceWindow(StWindow):
def __init__(
self, container: "DeltaGenerator" = st.container(), show_llm: bool = False, show_common_logs: bool = False
):
super().__init__(container)
self.show_llm = show_llm
self.show_common_logs = show_common_logs
self.pid_trace = ""
self.current_tag = ""
self.current_win = StWindow(self.container)
self.evolving_tasks: list[str] = []
def consume_msg(self, msg: Message):
# divide tag levels
if len(msg.tag) > len(self.current_tag):
# write a header about current task, if it is llm message, not write.
if not msg.tag.endswith("llm_messages"):
self.container.header(msg.tag.replace(".", " ➡ "), divider=True)
self.current_tag = msg.tag
# set log writer (window) according to msg
if msg.tag.endswith("llm_messages"):
# llm messages logs
if not self.show_llm:
return
if not isinstance(self.current_win, LLMWindow):
self.current_win = LLMWindow(self.container)
elif isinstance(msg.content, Hypothesis):
# hypothesis
self.current_win = HypothesisWindow(self.container)
elif isinstance(msg.content, HypothesisFeedback):
# hypothesis feedback
self.current_win = HypothesisFeedbackWindow(self.container)
elif isinstance(msg.content, QlibFactorExperiment):
self.current_win = QlibFactorExpWindow(self.container)
elif isinstance(msg.content, QlibModelExperiment):
self.current_win = QlibModelExpWindow(self.container)
elif isinstance(msg.content, list):
msg.content = [m for m in msg.content if m]
if len(msg.content) == 0:
return
if isinstance(msg.content[0], FactorTask):
self.current_win = ObjectsTabsWindow(
self.container.expander("Factor Tasks"), FactorTaskWindow, lambda x: x.factor_name
)
elif isinstance(msg.content[0], ModelTask):
self.current_win = ObjectsTabsWindow(
self.container.expander("Model Tasks"), ModelTaskWindow, lambda x: x.name
)
elif isinstance(msg.content[0], FactorFBWorkspace):
self.current_win = ObjectsTabsWindow(
self.container.expander("Factor Workspaces"),
inner_class=WorkspaceWindow,
mapper=lambda x: x.target_task.factor_name,
)
self.evolving_tasks = [m.target_task.factor_name for m in msg.content]
elif isinstance(msg.content[0], ModelFBWorkspace):
self.current_win = ObjectsTabsWindow(
self.container.expander("Model Workspaces"),
inner_class=WorkspaceWindow,
mapper=lambda x: x.target_task.name,
)
self.evolving_tasks = [m.target_task.name for m in msg.content]
elif isinstance(msg.content[0], FactorSingleFeedback):
self.current_win = ObjectsTabsWindow(
self.container.expander("Factor Feedbacks"),
inner_class=FactorFeedbackWindow,
tab_names=self.evolving_tasks,
)
elif isinstance(msg.content[0], ModelSingleFeedback):
self.current_win = ObjectsTabsWindow(
self.container.expander("Model Feedbacks"),
inner_class=ModelFeedbackWindow,
tab_names=self.evolving_tasks,
)
else:
# common logs
if not self.show_common_logs:
return
self.current_win = StWindow(self.container)
self.current_win.consume_msg(msg)
def mock_msg(obj) -> Message:
return Message(tag="mock", level="INFO", timestamp=datetime.now(), pid_trace="000", caller="mock", content=obj)
class TraceObjWindow(StWindow):
def __init__(self, container: "DeltaGenerator" = st.container()):
self.container = container
def consume_msg(self, msg: Message | Trace):
if isinstance(msg, Message):
trace: Trace = msg.content
else:
trace = msg
for id, (h, e, hf) in enumerate(trace.hist):
self.container.header(f"Trace History {id}", divider=True)
HypothesisWindow(self.container).consume_msg(mock_msg(h))
if isinstance(e, QlibFactorExperiment):
QlibFactorExpWindow(self.container).consume_msg(mock_msg(e))
else:
QlibModelExpWindow(self.container).consume_msg(mock_msg(e))
HypothesisFeedbackWindow(self.container).consume_msg(mock_msg(hf))
class ResearchWindow(StWindow):
def consume_msg(self, msg: Message):
if msg.tag.endswith("hypothesis generation"):
HypothesisWindow(self.container.container()).consume_msg(msg)
elif msg.tag.endswith("experiment generation"):
if isinstance(msg.content, list):
if isinstance(msg.content[0], FactorTask):
self.container.markdown("**Factor Tasks**")
ObjectsTabsWindow(
self.container.container(), FactorTaskWindow, lambda x: x.factor_name
).consume_msg(msg)
elif isinstance(msg.content[0], ModelTask):
self.container.markdown("**Model Tasks**")
ObjectsTabsWindow(self.container.container(), ModelTaskWindow, lambda x: x.name).consume_msg(msg)
elif msg.tag.endswith("load_pdf_screenshot"):
self.container.image(msg.content)
elif msg.tag.endswith("load_factor_tasks"):
self.container.json(msg.content)
class EvolvingWindow(StWindow):
def __init__(self, container: "DeltaGenerator"):
self.container = container
self.evolving_tasks: list[str] = []
def consume_msg(self, msg: Message):
if msg.tag.endswith("evolving code"):
if isinstance(msg.content, list):
msg.content = [m for m in msg.content if m]
if len(msg.content) == 0:
return
if isinstance(msg.content[0], FactorFBWorkspace):
self.container.markdown("**Factor Codes**")
ObjectsTabsWindow(
self.container.container(),
inner_class=WorkspaceWindow,
mapper=lambda x: x.target_task.factor_name,
).consume_msg(msg)
self.evolving_tasks = [m.target_task.factor_name for m in msg.content]
elif isinstance(msg.content[0], ModelFBWorkspace):
self.container.markdown("**Model Codes**")
ObjectsTabsWindow(
self.container.container(), inner_class=WorkspaceWindow, mapper=lambda x: x.target_task.name
).consume_msg(msg)
self.evolving_tasks = [m.target_task.name for m in msg.content]
elif msg.tag.endswith("evolving feedback"):
if isinstance(msg.content, list):
msg.content = [m for m in msg.content if m]
if len(msg.content) == 0:
return
if isinstance(msg.content[0], FactorSingleFeedback):
self.container.markdown("**Factor Feedbacks🔍**")
ObjectsTabsWindow(
self.container.container(), inner_class=FactorFeedbackWindow, tab_names=self.evolving_tasks
).consume_msg(msg)
elif isinstance(msg.content[0], ModelSingleFeedback):
self.container.markdown("**Model Feedbacks🔍**")
ObjectsTabsWindow(
self.container.container(), inner_class=ModelFeedbackWindow, tab_names=self.evolving_tasks
).consume_msg(msg)
class DevelopmentWindow(StWindow):
def __init__(self, container: "DeltaGenerator"):
self.E_win = RoundTabsWindow(
container.container(),
new_tab_func=lambda x: x.tag.endswith("evolving code"),
inner_class=EvolvingWindow,
title="Evolving Loops🔧",
)
def consume_msg(self, msg: Message):
if "evolving" in msg.tag:
self.E_win.consume_msg(msg)
class FeedbackWindow(StWindow):
def __init__(self, container: "DeltaGenerator"):
self.container = container
def consume_msg(self, msg: Message):
if msg.tag.endswith("returns"):
fig = px.line(msg.content)
self.container.markdown("**Returns📈**")
self.container.plotly_chart(fig)
elif isinstance(msg.content, HypothesisFeedback):
HypothesisFeedbackWindow(self.container.container(border=True)).consume_msg(msg)
elif isinstance(msg.content, QlibModelExperiment):
QlibModelExpWindow(self.container.container(border=True)).consume_msg(msg)
elif isinstance(msg.content, QlibFactorExperiment):
QlibFactorExpWindow(self.container.container(border=True)).consume_msg(msg)
class SingleRDLoopWindow(StWindow):
def __init__(self, container: "DeltaGenerator"):
self.container = container
col1, col2 = self.container.columns([2, 3])
self.R_win = ResearchWindow(col1.container(border=True))
self.F_win = FeedbackWindow(col1.container(border=True))
self.D_win = DevelopmentWindow(col2.container(border=True))
def consume_msg(self, msg: Message):
tags = msg.tag.split(".")
if "r" in tags:
self.R_win.consume_msg(msg)
elif "d" in tags:
self.D_win.consume_msg(msg)
elif "ef" in tags:
self.F_win.consume_msg(msg)
class TraceWindow(StWindow):
def __init__(
self, container: "DeltaGenerator" = st.container(), show_llm: bool = False, show_common_logs: bool = False
):
self.show_llm = show_llm
self.show_common_logs = show_common_logs
image_c, scen_c = container.columns([2, 3], vertical_alignment="center")
image_c.image("scen.png")
scen_c.container(border=True).markdown(QlibModelScenario().rich_style_description)
top_container = container.container()
col1, col2 = top_container.columns([2, 3])
chart_c = col2.container(border=True, height=500)
chart_c.markdown("**Metrics📈**")
self.chart_c = chart_c.empty()
hypothesis_status_c = col1.container(border=True, height=500)
hypothesis_status_c.markdown("**Hypotheses🏅**")
self.summary_c = hypothesis_status_c.empty()
self.RDL_win = RoundTabsWindow(
container.container(),
new_tab_func=lambda x: x.tag.endswith("hypothesis generation"),
inner_class=SingleRDLoopWindow,
title="R&D Loops♾️",
)
self.hypothesis_decisions = defaultdict(bool)
self.hypotheses: list[Hypothesis] = []
self.results = []
def consume_msg(self, msg: Message):
if not self.show_llm and "llm_messages" in msg.tag:
return
if not self.show_common_logs and isinstance(msg.content, str):
return
if isinstance(msg.content, dict):
return
if msg.tag.endswith("hypothesis generation"):
self.hypotheses.append(msg.content)
elif msg.tag.endswith("ef.feedback"):
self.hypothesis_decisions[self.hypotheses[-1]] = msg.content.decision
self.summary_c.markdown(
"\n".join(
(
f"{id+1}. :green[{self.hypotheses[id].hypothesis}]\n\t>*{self.hypotheses[id].concise_reason}*"
if d
else f"{id+1}. {self.hypotheses[id].hypothesis}\n\t>*{self.hypotheses[id].concise_reason}*"
)
for id, (h, d) in enumerate(self.hypothesis_decisions.items())
)
)
elif msg.tag.endswith("ef.model runner result") or msg.tag.endswith("ef.factor runner result"):
self.results.append(msg.content.result)
if len(self.results) == 1:
self.chart_c.table(self.results[0])
else:
df = pd.DataFrame(self.results, index=range(1, len(self.results) + 1))
fig = px.line(df, x=df.index, y=df.columns, markers=True)
self.chart_c.plotly_chart(fig)
self.RDL_win.consume_msg(msg)
# time.sleep(TIME_DELAY)
rdagent/log/utils/__init__.py
-129
View File
@@ -1,129 +0,0 @@
import inspect
import json
import re
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Optional, TypedDict, cast
class LogColors:
"""
ANSI color codes for use in console output.
"""
RED = "\033[91m"
GREEN = "\033[92m"
YELLOW = "\033[93m"
BLUE = "\033[94m"
MAGENTA = "\033[95m"
CYAN = "\033[96m"
WHITE = "\033[97m"
GRAY = "\033[90m"
BLACK = "\033[30m"
BOLD = "\033[1m"
ITALIC = "\033[3m"
END = "\033[0m"
@classmethod
def get_all_colors(cls: type["LogColors"]) -> list:
names = dir(cls)
names = [name for name in names if not name.startswith("__") and not callable(getattr(cls, name))]
return [getattr(cls, name) for name in names]
def render(self, text: str, color: str = "", style: str = "") -> str:
"""
render text by input color and style.
It's not recommend that input text is already rendered.
"""
# This method is called too frequently, which is not good.
colors = self.get_all_colors()
# Perhaps color and font should be distinguished here.
if color and color in colors:
error_message = f"color should be in: {colors} but now is: {color}"
raise ValueError(error_message)
if style and style in colors:
error_message = f"style should be in: {colors} but now is: {style}"
raise ValueError(error_message)
text = f"{color}{text}{self.END}"
return f"{style}{text}{self.END}"
@staticmethod
def remove_ansi_codes(s: str) -> str:
"""
It is for removing ansi ctrl characters in the string(e.g. colored text)
"""
ansi_escape = re.compile(r"\x1B\[[0-?]*[ -/]*[@-~]")
return ansi_escape.sub("", s)
class CallerInfo(TypedDict):
function: str
line: int
name: Optional[str]
def get_caller_info(level: int = 2) -> CallerInfo:
# Get the current stack information
stack = inspect.stack()
# The second element is usually the caller's information
caller_info = stack[level]
frame = caller_info[0]
info: CallerInfo = {
"line": caller_info.lineno,
"name": frame.f_globals["__name__"], # Get the module name from the frame's globals
"function": frame.f_code.co_name, # Get the caller's function name
}
return info
def is_valid_session(log_path: Path) -> bool:
return log_path.is_dir() and log_path.joinpath("__session__").exists()
def extract_loopid_func_name(tag: str) -> tuple[str, str] | tuple[None, None]:
"""extract loop id and function name from the tag in Message"""
match = re.search(r"Loop_(\d+)\.([^.]+)", tag)
return cast(tuple[str, str], match.groups()) if match else (None, None)
def extract_evoid(tag: str) -> str | None:
"""extract evo id from the tag in Message"""
match = re.search(r"evo_loop_(\d+)\.", tag)
return cast(str, match.group(1)) if match else None
def extract_json(log_content: str) -> dict | None:
match = re.search(r"\{.*\}", log_content, re.DOTALL)
if match:
return cast(dict, json.loads(match.group(0)))
return None
def gen_datetime(dt: datetime | None = None) -> datetime:
"""
Generate a datetime object in UTC timezone.
- If `dt` is None, it will return the current time in UTC.
- If `dt` is provided, it will convert it to UTC timezone.
"""
if dt is None:
return datetime.now(timezone.utc)
return dt.astimezone(timezone.utc)
def dict_get_with_warning(d: dict, key: str, default: Any = None) -> Any:
"""
Motivation:
- When handling the repsonse from the LLM, we may use dict get to get the value.
- the function prevent falling into default value **silently**.
- Instead, it will log a warning message.
"""
from rdagent.log import rdagent_logger as logger
if key not in d:
logger.warning(f"Key {key} not found in {d}")
return default
return d[key]
rdagent/log/utils/folder.py
-77
View File
@@ -1,77 +0,0 @@
"""
This module provides some useful functions for working with logger folders.
"""
import pickle
from datetime import timedelta
from pathlib import Path
import pandas as pd
from rdagent.utils.workflow import LoopBase
def get_first_session_file_after_duration(log_folder: str | Path, duration: str | pd.Timedelta) -> Path:
log_folder = Path(log_folder)
duration_dt = pd.Timedelta(duration)
# iterate the dump steps in increasing order
files = sorted(
(log_folder / "__session__").glob("*/*_*"), key=lambda f: (int(f.parent.name), int(f.name.split("_")[0]))
)
fp = None
for fp in files:
with fp.open("rb") as f:
session_obj: LoopBase = pickle.load(f)
timer = session_obj.timer
all_duration = timer.all_duration
remain_time_duration = timer.remain_time()
if all_duration is None or remain_time_duration is None:
msg = "Timer is not configured"
raise ValueError(msg)
time_spent = all_duration - remain_time_duration
if time_spent >= duration_dt:
break
if fp is None:
msg = f"No session file found after duration {duration}"
raise ValueError(msg)
return fp
def first_li_si_after_one_time(log_path: Path, hours: int = 12) -> tuple[int, int, str]:
"""
Based on the hours, find the stop loop id and step id (the first step after <hours> hours).
Args:
log_path (Path): The path to the log folder (contains many log traces).
hours (int): The number of hours to stat.
Returns:
tuple[int, int, str]: The loop id, step id and function name.
"""
session_path = log_path / "__session__"
max_li = max(int(p.name) for p in session_path.iterdir() if p.is_dir() and p.name.isdigit())
max_step = max(int(p.name.split("_")[0]) for p in (session_path / str(max_li)).iterdir() if p.is_file())
rdloop_obj_p = next((session_path / str(max_li)).glob(f"{max_step}_*"))
rdloop_obj = DataScienceRDLoop.load(rdloop_obj_p)
loop_trace = rdloop_obj.loop_trace
si2fn = rdloop_obj.steps
duration = timedelta(seconds=0)
for li, lts in loop_trace.items():
for lt in lts:
si = lt.step_idx
duration += lt.end - lt.start
if duration > timedelta(hours=hours):
return li, si, si2fn[si]
if __name__ == "__main__":
from rdagent.app.data_science.loop import DataScienceRDLoop
f = get_first_session_file_after_duration("<path to log aptos2019-blindness-detection>", pd.Timedelta("12h"))
with f.open("rb") as f:
session_obj: LoopBase = pickle.load(f)
loop_trace = session_obj.loop_trace
last_loop = loop_trace[max(loop_trace.keys())]
last_step = last_loop[-1]
session_obj.steps[last_step.step_idx]
setup_predix_eurusd.sh
Executable
+301
View File
@@ -0,0 +1,301 @@
#!/bin/bash
# =============================================================================
# setup_predix_eurusd.sh
# Richtet Predix für EURUSD 15min Trading ein
# Ausführen: bash setup_predix_eurusd.sh
# =============================================================================
set -e # Abbruch bei Fehler
PREDIX_DIR="$HOME/Predix"
CSV_SOURCE="$HOME/Downloads/eurusd_data.csv"
DATA_DIR="$PREDIX_DIR/git_ignore_folder/eurusd_data"
QLIB_DIR="$HOME/.qlib/qlib_data/eurusd_data"
echo "========================================"
echo " Predix EURUSD Setup"
echo "========================================"
# ─── 1. Prüfen ob alles da ist ───────────────────────────────────────────────
echo ""
echo "[1/7] Prüfe Voraussetzungen..."
if [ ! -d "$PREDIX_DIR" ]; then
echo "FEHLER: $PREDIX_DIR nicht gefunden!"
exit 1
fi
if [ ! -f "$CSV_SOURCE" ]; then
echo "FEHLER: $CSV_SOURCE nicht gefunden!"
echo "Bitte eurusd_data.csv in ~/Downloads/ legen"
exit 1
fi
echo "✓ Predix gefunden: $PREDIX_DIR"
echo "✓ CSV gefunden: $CSV_SOURCE"
# ─── 2. Ordner anlegen ───────────────────────────────────────────────────────
echo ""
echo "[2/7] Erstelle Ordnerstruktur..."
mkdir -p "$DATA_DIR"
mkdir -p "$QLIB_DIR/calendars"
mkdir -p "$QLIB_DIR/instruments"
mkdir -p "$QLIB_DIR/features/eurusd"
mkdir -p "$PREDIX_DIR/git_ignore_folder/log"
cp "$CSV_SOURCE" "$DATA_DIR/eurusd_data.csv"
echo "✓ CSV kopiert nach $DATA_DIR"
# ─── 3. CSV → Qlib Format konvertieren ───────────────────────────────────────
echo ""
echo "[3/7] Konvertiere CSV zu Qlib-Format..."
python3 << 'PYEOF'
import pandas as pd
import numpy as np
from pathlib import Path
import os
QLIB_DIR = Path(os.path.expanduser("~/.qlib/qlib_data/eurusd_data"))
CSV_PATH = Path(os.path.expanduser("~/Downloads/eurusd_data.csv"))
# Laden + sortieren
df = pd.read_csv(CSV_PATH, parse_dates=["datetime"])
df = df.sort_values("datetime").reset_index(drop=True)
df.columns = [c.lower() for c in df.columns]
print(f" Rows: {len(df):,} | Range: {df['datetime'].min().date()} -> {df['datetime'].max().date()}")
# ── Kalender (alle 15min Timestamps) ────────────────────────────────────────
cal = df["datetime"].dt.strftime("%Y-%m-%d %H:%M:%S")
cal_path = QLIB_DIR / "calendars" / "15min.txt"
cal.to_csv(cal_path, index=False, header=False)
print(f" ✓ Kalender: {len(cal)} Einträge -> {cal_path}")
# ── Instruments (nur EURUSD) ─────────────────────────────────────────────────
inst_path = QLIB_DIR / "instruments" / "all.txt"
start = df["datetime"].min().strftime("%Y-%m-%d")
end = df["datetime"].max().strftime("%Y-%m-%d")
with open(inst_path, "w") as f:
f.write(f"EURUSD\t{start}\t{end}\n")
print(f" ✓ Instruments -> {inst_path}")
# ── Features (Qlib binary format via CSV) ───────────────────────────────────
feat_dir = QLIB_DIR / "features" / "eurusd"
feat_dir.mkdir(parents=True, exist_ok=True)
# Qlib erwartet: $open, $high, $low, $close, $volume
for col in ["open", "high", "low", "close", "volume"]:
out = feat_dir / f"{col}.day.bin"
# Qlib binary: float32 array
arr = df[col].astype("float32").values
arr.tofile(str(out).replace(".day.bin", f"_15min.bin"))
# Auch als einfache CSV für direkten Zugriff
df.to_csv(QLIB_DIR / "eurusd_15min.csv", index=False)
print(f" ✓ Features + CSV -> {feat_dir}")
# ── Returns + technische Features vorberechnen ───────────────────────────────
def ema(s, p): return s.ewm(span=p, adjust=False).mean()
def rsi(c, p=14):
d = c.diff()
g = d.clip(lower=0).ewm(span=p, adjust=False).mean()
l = (-d.clip(upper=0)).ewm(span=p, adjust=False).mean()
return 100 - 100/(1 + g/(l+1e-9))
feat = pd.DataFrame()
feat["datetime"] = df["datetime"]
feat["close"] = df["close"]
for n in [1,4,8,16,96]:
feat[f"ret_{n}"] = df["close"].pct_change(n)
feat["rsi_14"] = rsi(df["close"], 14)
feat["macd_hist"] = ema(df["close"],12) - ema(df["close"],26)
feat["hour"] = df["datetime"].dt.hour
feat["is_london"] = feat["hour"].isin([8,9,10,11]).astype(int)
feat["is_ny"] = feat["hour"].isin([13,14,15,16]).astype(int)
feat["adx_proxy"] = df["close"].rolling(14).std() / df["close"].rolling(96).std()
feat.dropna(inplace=True)
feat.to_csv(QLIB_DIR / "eurusd_features.csv", index=False)
print(f" ✓ Features CSV: {len(feat):,} Zeilen, {len(feat.columns)} Spalten")
print(" Fertig!")
PYEOF
echo "✓ Qlib-Daten konvertiert"
# ─── 4. .env updaten ─────────────────────────────────────────────────────────
echo ""
echo "[4/7] Update .env..."
ENV_FILE="$PREDIX_DIR/.env"
# Backup
cp "$ENV_FILE" "$ENV_FILE.backup_$(date +%Y%m%d_%H%M%S)"
# QLIB_DATA_DIR auf EURUSD umbiegen
sed -i "s|QLIB_DATA_DIR=.*|QLIB_DATA_DIR=$QLIB_DIR|" "$ENV_FILE"
# LOG_PATH korrigieren (war /home/nico/RD-Agent-Local/log)
sed -i "s|LOG_PATH=.*|LOG_PATH=$PREDIX_DIR/git_ignore_folder/log|" "$ENV_FILE"
# EURUSD-spezifische Vars hinzufügen (falls noch nicht da)
grep -q "EURUSD_DATA_PATH" "$ENV_FILE" || cat >> "$ENV_FILE" << 'ENVEOF'
# ---------- EURUSD ----------
EURUSD_DATA_PATH=/home/nico/.qlib/qlib_data/eurusd_data/eurusd_15min.csv
QLIB_FREQ=15min
QLIB_MARKET=eurusd
BACKTEST_START_TIME=2024-08-09
BACKTEST_END_TIME=2026-03-20
COST_RATE=0.00015
ENVEOF
echo "✓ .env aktualisiert (Backup erstellt)"
# ─── 5. Prompts für EURUSD anpassen ──────────────────────────────────────────
echo ""
echo "[5/7] Passe Qlib-Prompts für EURUSD an..."
PROMPT_FILE="$PREDIX_DIR/rdagent/app/qlib_rd_loop/prompts.yaml"
cp "$PROMPT_FILE" "${PROMPT_FILE}.backup_$(date +%Y%m%d_%H%M%S)"
cat > "$PROMPT_FILE" << 'YAMLEOF'
hypothesis_generation:
system: |-
You are an expert quantitative researcher specialized in FX (foreign exchange) trading,
specifically EURUSD intraday strategies on 15-minute bars.
EURUSD domain knowledge you must apply:
- London session (08:00-12:00 UTC): highest volatility, trending behavior — favor momentum strategies
- NY session (13:00-17:00 UTC): second volatility peak, also trending
- Asian session (00:00-07:00 UTC): low volatility, mean-reverting behavior
- London/NY overlap (13:00-17:00 UTC): strongest directional moves of the day
- Weekend gap risk: avoid holding positions after Friday 20:00 UTC
- Spread cost: ~1.5 bps per trade — strategies must minimize unnecessary entries
- EURUSD is mean-reverting on short windows (<1h), trending on longer (>4h)
- Key macro drivers: ECB/Fed rate decisions, NFP (first Friday of month), CPI releases
Available model types you can propose:
- TimeSeries: LSTM, GRU, TCN (Temporal Convolutional Network), Transformer, PatchTST
- Tabular: XGBoost, LightGBM, RandomForest (on engineered features)
- Hybrid: CNN+LSTM, XGBoost+LSTM ensemble
- Statistical: Regime-switching (HMM), Kalman filter
Available features in the dataset:
- OHLCV: open, high, low, close, volume (15min bars)
- Returns: ret_1, ret_4, ret_8, ret_16, ret_96
- Technical: rsi_14, macd_hist, adx_14, atr_14, bb_pct, stoch_k, cci_14
- Volatility: vol_real_4, vol_real_16, vol_ratio, zscore_ret_96
- Time/Session: hour, is_london, is_ny, is_overlap, hour_sin, hour_cos
- Lags: rsi_14_lag1-8, macd_hist_lag1-8, bb_pct_lag1-8
Your hypothesis must:
1. Specify which session(s) the strategy targets
2. Name which model type to use and why it fits EURUSD
3. Include a session filter (is_london / is_ny)
4. Include a spread filter (only trade when expected |return| > 0.0003)
5. Specify target: classification (fwd_sign_4) or regression (fwd_ret_4)
Please ensure your response is in JSON format:
{
"hypothesis": "A clear and concise trading hypothesis for EURUSD 15min.",
"reason": "Detailed explanation including session, model choice, and expected edge.",
"model_type": "One of: TimeSeries / Tabular / XGBoost",
"target_session": "london / ny / asian / all",
"expected_arr_range": "e.g. 8-12%"
}
user: |-
Previously tried approaches and their results:
{{ factor_descriptions }}
Additional context:
{{ report_content }}
Generate a NEW hypothesis that is meaningfully different from what has been tried.
Focus on approaches that have NOT been tested yet.
Target: beat current best ARR of 9.62%.
YAMLEOF
echo "✓ prompts.yaml aktualisiert (Backup erstellt)"
# ─── 6. Model Coder Prompt erweitern ─────────────────────────────────────────
echo ""
echo "[6/7] Erweitere Model Coder Prompts..."
MODEL_PROMPT="$PREDIX_DIR/rdagent/components/coder/model_coder/prompts.yaml"
cp "$MODEL_PROMPT" "${MODEL_PROMPT}.backup_$(date +%Y%m%d_%H%M%S)"
# EURUSD session filter als Kommentar in den evolving_strategy Block injizieren
python3 << 'PYEOF'
import re
from pathlib import Path
path = Path("/home/nico/Predix/rdagent/components/coder/model_coder/prompts.yaml")
content = path.read_text()
eurusd_note = """
EURUSD-specific rules (ALWAYS apply these in generated code):
1. Session filter: use is_london and is_ny columns — weight/filter signals to active sessions
2. Spread filter: only generate signal when abs(predicted_return) > 0.0003
3. ADX regime: if adx_proxy > 1.2 use trend model; if adx_proxy < 0.8 use mean-reversion
4. Weekend filter: zero out signals when dayofweek==4 and hour>=20
5. Max trade frequency: target <15 trades per day (avoid spread cost death)
6. Supported model_type values: "Tabular", "TimeSeries", "XGBoost"
"""
# Inject after the scenario line in evolving_strategy_model_coder
content = content.replace(
" Your code is expected to align the scenario in any form",
eurusd_note + "\n Your code is expected to align the scenario in any form"
)
path.write_text(content)
print(" ✓ Model coder prompt erweitert")
PYEOF
echo "✓ Model coder Prompt angepasst"
# ─── 7. Git Commits ──────────────────────────────────────────────────────────
echo ""
echo "[7/7] Git Commits..."
cd "$PREDIX_DIR"
git add rdagent/app/qlib_rd_loop/prompts.yaml
git commit -m "feat: EURUSD 15min prompts - session filter, FX domain knowledge
- Add London/NY/Asian session awareness to hypothesis generation
- Add model type suggestions: LSTM, GRU, TCN, Transformer, XGBoost, LightGBM
- Add spread filter (1.5 bps) and ADX regime detection
- Target: beat current best ARR of 9.62%"
git add rdagent/components/coder/model_coder/prompts.yaml
git commit -m "feat: inject EURUSD trading rules into model coder
- Session filter (is_london, is_ny)
- Spread filter (|return| > 0.0003)
- Weekend position close
- Max trade frequency guidance"
git add .env 2>/dev/null || true
echo " (Note: .env nicht committed - enthält API Keys)"
echo ""
echo "========================================"
echo " Setup abgeschlossen!"
echo "========================================"
echo ""
echo "Nächste Schritte:"
echo ""
echo " 1. Starten:"
echo " cd ~/Predix && rdagent fin_quant"
echo ""
echo " 2. Dashboard (in zweitem Terminal):"
echo " cd ~/Predix && rdagent server_ui --port 19899"
echo " → http://localhost:19899"
echo ""
echo " 3. Logs:"
echo " tail -f $PREDIX_DIR/git_ignore_folder/log/*.log"
echo ""
echo "Backup-Dateien (.backup_*) können nach erfolgreichem Test gelöscht werden."