gavindiaz/NexQuant
1
0
Fork 0
mirror of https://github.com/NicolasBohn/NexQuant.git synced 2026-07-27 23:47:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor: remove unused code and update documentation in CoSTEER module (#720)

Browse Source 
* refactor: Remove unused code and update documentation in CoSTEER module

* fix mypy error

* fix black error

---------

Co-authored-by: Young <afe.young@gmail.com>
This commit is contained in:
Linlang
2025-03-26 22:48:01 +08:00
committed by GitHub
parent 525b1a2136
commit 6da6fb5ee5
12 changed files with 27 additions and 58 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/scens/data_agent_fin.rst
+1 -1
View File
@@ -133,6 +133,6 @@ The following environment variables can be set in the `.env` file to customize t
.. autopydantic_settings:: rdagent.components.coder.factor_coder.config.FactorCoSTEERSettings
:settings-show-field-summary: False
:members: coder_use_cache, data_folder, data_folder_debug, file_based_execution_timeout, select_method, select_threshold, max_loop, knowledge_base_path, new_knowledge_base_path
:members: coder_use_cache, data_folder, data_folder_debug, file_based_execution_timeout, select_method, max_loop, knowledge_base_path, new_knowledge_base_path
:exclude-members: Config, fail_task_trial_limit, v1_query_former_trace_limit, v1_query_similar_success_limit, v2_query_component_limit, v2_query_error_limit, v2_query_former_trace_limit, v2_error_summary, v2_knowledge_sampler
:no-index:
docs/scens/data_copilot_fin.rst
+1 -1
View File
@@ -159,6 +159,6 @@ The following environment variables can be set in the `.env` file to customize t
.. autopydantic_settings:: rdagent.components.coder.factor_coder.config.FactorCoSTEERSettings
:settings-show-field-summary: False
:members: coder_use_cache, data_folder, data_folder_debug, file_based_execution_timeout, select_method, select_threshold, max_loop, knowledge_base_path, new_knowledge_base_path
:members: coder_use_cache, data_folder, data_folder_debug, file_based_execution_timeout, select_method, max_loop, knowledge_base_path, new_knowledge_base_path
:exclude-members: Config, python_bin, fail_task_trial_limit, v1_query_former_trace_limit, v1_query_similar_success_limit, v2_query_component_limit, v2_query_error_limit, v2_query_former_trace_limit, v2_error_summary, v2_knowledge_sampler
:no-index:
docs/scens/kaggle_agent.rst
+1 -1
View File
@@ -268,5 +268,5 @@ The following environment variables can be set in the `.env` file to customize t
.. autopydantic_settings:: rdagent.components.coder.factor_coder.config.FactorCoSTEERSettings
:settings-show-field-summary: False
:members: coder_use_cache, file_based_execution_timeout, select_method, max_loop
:exclude-members: Config, fail_task_trial_limit, v1_query_former_trace_limit, v1_query_similar_success_limit, v2_query_component_limit, v2_query_error_limit, v2_query_former_trace_limit, v2_error_summary, v2_knowledge_sampler, v2_add_fail_attempt_to_latest_successful_execution, new_knowledge_base_path, knowledge_base_path, data_folder, data_folder_debug, select_threshold
:exclude-members: Config, fail_task_trial_limit, v1_query_former_trace_limit, v1_query_similar_success_limit, v2_query_component_limit, v2_query_error_limit, v2_query_former_trace_limit, v2_error_summary, v2_knowledge_sampler, v2_add_fail_attempt_to_latest_successful_execution, new_knowledge_base_path, knowledge_base_path, data_folder, data_folder_debug
:no-index:
rdagent/components/coder/CoSTEER/config.py
-2
View File
@@ -33,8 +33,6 @@ class CoSTEERSettings(ExtendedBaseSettings):
new_knowledge_base_path: Union[str, None] = None
"""Path to the new knowledge base"""
select_threshold: int = 10
max_seconds: int = 10**6
rdagent/components/coder/CoSTEER/evaluators.py
+8 -1
View File
@@ -162,7 +162,14 @@ class CoSTEERMultiFeedback(Feedback):
def __iter__(self):
return iter(self.feedback_list)
def __bool__(self):
def finished(self) -> bool:
"""
In some implementations, tasks may fail multiple times, leading agents to skip the implementation.
This results in None feedback. However, we want to accept the correct parts and ignore None feedback.
"""
return all(feedback.final_decision for feedback in self.feedback_list if feedback is not None)
def __bool__(self) -> bool:
return all(feedback.final_decision for feedback in self.feedback_list)
rdagent/components/coder/CoSTEER/evolvable_subjects.py
-1
View File
@@ -14,7 +14,6 @@ class EvolvingItem(Experiment, EvolvableSubjects):
sub_gt_implementations: list[FBWorkspace] = None,
):
Experiment.__init__(self, sub_tasks=sub_tasks)
self.corresponding_selection: list = None
if sub_gt_implementations is not None and len(
sub_gt_implementations,
) != len(self.sub_tasks):
rdagent/components/coder/CoSTEER/evolving_strategy.py
+4 -23
View File
@@ -12,7 +12,6 @@ from rdagent.components.coder.CoSTEER.evolvable_subjects import EvolvingItem
from rdagent.components.coder.CoSTEER.knowledge_management import (
CoSTEERQueriedKnowledge,
)
from rdagent.components.coder.CoSTEER.scheduler import random_select
from rdagent.core.conf import RD_AGENT_SETTINGS
from rdagent.core.evolving_framework import EvolvingStrategy, EvoStep, QueriedKnowledge
from rdagent.core.experiment import FBWorkspace, Task
@@ -59,17 +58,6 @@ class MultiProcessEvolvingStrategy(EvolvingStrategy):
"""
raise NotImplementedError
def select_one_round_tasks(
self,
to_be_finished_task_index: list,
evo: EvolvingItem,
selected_num: int,
queried_knowledge: CoSTEERQueriedKnowledge,
scen: Scenario,
) -> list:
"""Since scheduler is not essential, we implement a simple random selection here."""
return random_select(to_be_finished_task_index, evo, selected_num, queried_knowledge, scen)
@abstractmethod
def assign_code_list_to_evo(self, code_list: list[dict], evo: EvolvingItem) -> None:
"""
@@ -92,10 +80,12 @@ class MultiProcessEvolvingStrategy(EvolvingStrategy):
**kwargs,
) -> EvolvingItem:
# 1.找出需要evolve的task
to_be_finished_task_index = []
to_be_finished_task_index: list[int] = []
for index, target_task in enumerate(evo.sub_tasks):
target_task_desc = target_task.get_task_information()
if target_task_desc in queried_knowledge.success_task_to_knowledge_dict:
# NOTE: very weird logic:
# it depends on the knowledge to set the already finished task
evo.sub_workspace_list[index] = queried_knowledge.success_task_to_knowledge_dict[
target_task_desc
].implementation
@@ -105,19 +95,11 @@ class MultiProcessEvolvingStrategy(EvolvingStrategy):
):
to_be_finished_task_index.append(index)
# 2. 选择selection方法
# if the number of factors to be implemented is larger than the limit, we need to select some of them
if self.settings.select_threshold < len(to_be_finished_task_index):
# Select a fixed number of factors if the total exceeds the threshold
to_be_finished_task_index = self.select_one_round_tasks(
to_be_finished_task_index, evo, self.settings.select_threshold, queried_knowledge, self.scen
)
last_feedback = None
if len(evolving_trace) > 0:
last_feedback = evolving_trace[-1].feedback
assert isinstance(last_feedback, CoSTEERMultiFeedback)
result = multiprocessing_wrapper(
[
(
@@ -138,6 +120,5 @@ class MultiProcessEvolvingStrategy(EvolvingStrategy):
code_list[target_index] = result[index]
evo = self.assign_code_list_to_evo(code_list, evo)
evo.corresponding_selection = to_be_finished_task_index
return evo
rdagent/components/coder/CoSTEER/scheduler.py
-25
View File
@@ -1,25 +0,0 @@
import random
from rdagent.components.coder.CoSTEER.evolvable_subjects import EvolvingItem
from rdagent.components.coder.CoSTEER.knowledge_management import (
CoSTEERQueriedKnowledge,
)
from rdagent.core.scenario import Scenario
from rdagent.log import rdagent_logger as logger
def random_select(
to_be_finished_task_index: list,
evo: EvolvingItem,
selected_num: int,
queried_knowledge: CoSTEERQueriedKnowledge,
scen: Scenario,
):
to_be_finished_task_index = random.sample(
to_be_finished_task_index,
selected_num,
)
logger.info(f"The random selection is: {to_be_finished_task_index}")
return to_be_finished_task_index
rdagent/components/coder/factor_coder/evaluators.py
-1
View File
@@ -15,7 +15,6 @@ from rdagent.core.evolving_framework import QueriedKnowledge
from rdagent.core.experiment import Workspace
FactorSingleFeedback = CoSTEERSingleFeedbackDeprecated
FactorMultiFeedback = CoSTEERMultiFeedback
class FactorEvaluatorForCoder(CoSTEEREvaluator):
rdagent/core/evaluation.py
+7
View File
@@ -12,6 +12,13 @@ class Feedback:
The building process of feedback will should be in evaluator
"""
def finished(self) -> bool:
"""
In some implementations, tasks may fail multiple times, leading agents to skip the implementation.
So both skip and success indicate the task is finished.
"""
return self.__bool__()
def __bool__(self) -> bool:
return True
rdagent/core/evolving_agent.py
+1 -1
View File
@@ -102,6 +102,6 @@ class RAGEvoAgent(EvoAgent[RAGEvaluator]):
yield evo # yield the control to caller for process control and logging.
# 7. check if all tasks are completed
if self.with_feedback and es.feedback:
if self.with_feedback and es.feedback is not None and es.feedback.finished():
logger.info("All tasks in evolving subject have been completed.")
break
rdagent/core/experiment.py
+4 -1
View File
@@ -10,7 +10,7 @@ from abc import ABC, abstractmethod
from collections.abc import Sequence
from copy import deepcopy
from pathlib import Path
from typing import Any, Generic, TypeVar
from typing import Any, Generic, Literal, TypeVar
from rdagent.core.conf import RD_AGENT_SETTINGS
from rdagent.core.evaluation import Feedback
@@ -297,6 +297,9 @@ class Experiment(
) -> None:
self.hypothesis: Hypothesis | None = hypothesis # Experiment is optionally generated by hypothesis
self.sub_tasks: Sequence[ASpecificTask] = sub_tasks
# None means
# - initialization placeholder before implementation
# - the developer actively skip the task;
self.sub_workspace_list: list[ASpecificWSForSubTasks | None] = [None] * len(self.sub_tasks)
# TODO:
# It will be used in runner in history