import argparse import textwrap from collections import defaultdict from datetime import datetime, timezone from pathlib import Path from typing import Callable, Type import pandas as pd import plotly.express as px import plotly.graph_objects as go import streamlit as st from plotly.subplots import make_subplots from st_btn_select import st_btn_select from streamlit import session_state as state from streamlit.delta_generator import DeltaGenerator from rdagent.app.model_extraction_and_code.GeneralModel import GeneralModelScenario from rdagent.components.coder.factor_coder.CoSTEER.evaluators import ( FactorSingleFeedback, ) from rdagent.components.coder.factor_coder.factor import FactorFBWorkspace, FactorTask from rdagent.components.coder.model_coder.CoSTEER.evaluators import ModelCoderFeedback from rdagent.components.coder.model_coder.model import ModelFBWorkspace, ModelTask from rdagent.core.proposal import Hypothesis, HypothesisFeedback from rdagent.log.base import Message from rdagent.log.storage import FileStorage from rdagent.log.ui.qlib_report_figure import report_figure from rdagent.scenarios.data_mining.experiment.model_experiment import DMModelScenario from rdagent.scenarios.qlib.experiment.factor_experiment import ( QlibFactorExperiment, QlibFactorScenario, ) from rdagent.scenarios.qlib.experiment.model_experiment import ( QlibModelExperiment, QlibModelScenario, ) st.set_page_config(layout="wide", page_title="RD-Agent", 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() if args.log_dir: main_log_path = Path(args.log_dir) if not main_log_path.exists(): st.error(f"Log dir `{main_log_path}` does not exist!") st.stop() else: main_log_path = None SELECTED_METRICS = [ "IC", "1day.excess_return_without_cost.annualized_return", "1day.excess_return_without_cost.information_ratio", "1day.excess_return_without_cost.max_drawdown", ] if "log_type" not in state: state.log_type = "Qlib Model" if "log_path" not in state: if main_log_path: state.log_path = next(main_log_path.iterdir()).relative_to(main_log_path) else: state.log_path = "" if "fs" not in state: state.fs = None if "msgs" not in state: state.msgs = defaultdict(lambda: defaultdict(list)) if "last_msg" not in state: state.last_msg = None if "current_tags" not in state: state.current_tags = [] if "lround" not in state: state.lround = 0 # RD Loop Round if "erounds" not in state: state.erounds = defaultdict(int) # Evolving Rounds in each RD Loop if "e_decisions" not in state: state.e_decisions = defaultdict(lambda: defaultdict(tuple)) # Summary Info if "hypotheses" not in state: # Hypotheses in each RD Loop state.hypotheses = defaultdict(None) if "h_decisions" not in state: state.h_decisions = defaultdict(bool) if "metric_series" not in state: state.metric_series = [] # Factor Task Baseline if "alpha158_metrics" not in state: state.alpha158_metrics = None def refresh(): if main_log_path: state.fs = FileStorage(main_log_path / state.log_path).iter_msg() else: state.fs = FileStorage(state.log_path).iter_msg() state.msgs = defaultdict(lambda: defaultdict(list)) state.lround = 0 state.erounds = defaultdict(int) state.e_decisions = defaultdict(lambda: defaultdict(tuple)) state.hypotheses = defaultdict(None) state.h_decisions = defaultdict(bool) state.metric_series = [] state.last_msg = None state.current_tags = [] state.alpha158_metrics = None def should_display(msg: Message): for t in state.excluded_tags: if t in msg.tag.split("."): return False if type(msg.content).__name__ in state.excluded_types: return False return True def get_msgs_until(end_func: Callable[[Message], bool] = lambda _: True): if state.fs: while True: try: msg = next(state.fs) if should_display(msg): tags = msg.tag.split(".") if "r" not in state.current_tags and "r" in tags: state.lround += 1 if "evolving code" not in state.current_tags and "evolving code" in tags: state.erounds[state.lround] += 1 state.current_tags = tags state.last_msg = msg # Update Summary Info if "model runner result" in tags or "factor runner result" in tags or "runner result" in tags: # factor baseline exp metrics if state.log_type == "Qlib Factor" and state.alpha158_metrics is None: sms = msg.content.based_experiments[0].result.loc[SELECTED_METRICS] sms.name = "alpha158" state.alpha158_metrics = sms # common metrics if msg.content.result is None: state.metric_series.append(pd.Series([None], index=["AUROC"], name=f"Round {state.lround}")) else: if len(msg.content.result) < 4: ps = msg.content.result ps.index = ["AUROC"] ps.name = f"Round {state.lround}" state.metric_series.append(ps) else: sms = msg.content.result.loc[SELECTED_METRICS] sms.name = f"Round {state.lround}" state.metric_series.append(sms) elif "hypothesis generation" in tags: state.hypotheses[state.lround] = msg.content elif "ef" in tags and "feedback" in tags: state.h_decisions[state.lround] = msg.content.decision elif "d" in tags: if "evolving code" in tags: msg.content = [i for i in msg.content if i] if "evolving feedback" in tags: msg.content = [i for i in msg.content if i] if len(msg.content) != len(state.msgs[state.lround]["d.evolving code"][-1].content): st.toast(":red[**Evolving Feedback Length Error!**]", icon="‼️") right_num = 0 for wsf in msg.content: if wsf.final_decision: right_num += 1 wrong_num = len(msg.content) - right_num state.e_decisions[state.lround][state.erounds[state.lround]] = (right_num, wrong_num) state.msgs[state.lround][msg.tag].append(msg) # Stop Getting Logs if end_func(msg): break except StopIteration: st.toast(":red[**No More Logs to Show!**]", icon="🛑") break def evolving_feedback_window(wsf: FactorSingleFeedback | ModelCoderFeedback): if isinstance(wsf, FactorSingleFeedback): ffc, efc, cfc, vfc = st.tabs( ["**Final Feedback🏁**", "Execution Feedback🖥️", "Code Feedback📄", "Value Feedback🔢"] ) with ffc: st.markdown(wsf.final_feedback) with efc: st.code(wsf.execution_feedback, language="log") with cfc: st.markdown(wsf.code_feedback) with vfc: st.markdown(wsf.factor_value_feedback) elif isinstance(wsf, ModelCoderFeedback): ffc, efc, cfc, msfc, vfc = st.tabs( [ "**Final Feedback🏁**", "Execution Feedback🖥️", "Code Feedback📄", "Model Shape Feedback📐", "Value Feedback🔢", ] ) with ffc: st.markdown(wsf.final_feedback) with efc: st.code(wsf.execution_feedback, language="log") with cfc: st.markdown(wsf.code_feedback) with msfc: st.markdown(wsf.shape_feedback) with vfc: st.markdown(wsf.value_feedback) def display_hypotheses(hypotheses: dict[int, Hypothesis], decisions: dict[int, bool], success_only: bool = False): if success_only: shd = {k: v.__dict__ for k, v in hypotheses.items() if decisions[k]} else: shd = {k: v.__dict__ for k, v in hypotheses.items()} df = pd.DataFrame(shd).T if "reason" in df.columns: df.drop(["reason"], axis=1, inplace=True) df.columns = df.columns.map(lambda x: x.replace("_", " ").capitalize()) def style_rows(row): if decisions[row.name]: return ["color: green;"] * len(row) return [""] * len(row) def style_columns(col): if col.name != "Hypothesis": return ["font-style: italic;"] * len(col) return ["font-weight: bold;"] * len(col) # st.dataframe(df.style.apply(style_rows, axis=1).apply(style_columns, axis=0)) st.markdown(df.style.apply(style_rows, axis=1).apply(style_columns, axis=0).to_html(), unsafe_allow_html=True) def metrics_window(df: pd.DataFrame, R: int, C: int, *, height: int = 300, colors: list[str] = None): fig = make_subplots(rows=R, cols=C, subplot_titles=df.columns) def hypothesis_hover_text(h: Hypothesis, d: bool = False): color = "green" if d else "black" text = h.hypothesis lines = textwrap.wrap(text, width=60) return f"{'
'.join(lines)}" hover_texts = [ hypothesis_hover_text(state.hypotheses[int(i[6:])], state.h_decisions[int(i[6:])]) for i in df.index if i != "alpha158" ] if state.alpha158_metrics is not None: hover_texts = ["Baseline: alpha158"] + hover_texts for ci, col in enumerate(df.columns): row = ci // C + 1 col_num = ci % C + 1 fig.add_trace( go.Scatter( x=df.index, y=df[col], name=col, mode="lines+markers", connectgaps=True, marker=dict(size=10, color=colors[ci]) if colors else dict(size=10), hovertext=hover_texts, hovertemplate="%{hovertext}

%{x} Value: %{y}", ), row=row, col=col_num, ) fig.update_layout(showlegend=False, height=height) if state.alpha158_metrics is not None: for i in range(1, R + 1): # 行 for j in range(1, C + 1): # 列 fig.update_xaxes( tickvals=[df.index[0]] + list(df.index[1:]), ticktext=[f'{df.index[0]}'] + list(df.index[1:]), row=i, col=j, ) st.plotly_chart(fig) def summary_window(): if state.log_type in ["Qlib Model", "Data Mining", "Qlib Factor"]: st.header("Summary📊", divider="rainbow", anchor="_summary") with st.container(): # TODO: not fixed height with st.container(): bc, cc = st.columns([2, 2], vertical_alignment="center") with bc: st.subheader("Metrics📈", anchor="_metrics") with cc: show_true_only = st.toggle("successful hypotheses", value=False) # hypotheses_c, chart_c = st.columns([2, 3]) chart_c = st.container() hypotheses_c = st.container() with hypotheses_c: st.subheader("Hypotheses🏅", anchor="_hypotheses") display_hypotheses(state.hypotheses, state.h_decisions, show_true_only) with chart_c: if state.log_type == "Qlib Factor" and state.alpha158_metrics is not None: df = pd.DataFrame([state.alpha158_metrics] + state.metric_series) else: df = pd.DataFrame(state.metric_series) if show_true_only and len(state.hypotheses) >= len(state.metric_series): if state.alpha158_metrics is not None: selected = ["alpha158"] + [i for i in df.index if state.h_decisions[int(i[6:])]] else: selected = [i for i in df.index if state.h_decisions[int(i[6:])]] df = df.loc[selected] if df.shape[0] == 1: st.table(df.iloc[0]) elif df.shape[0] > 1: if df.shape[1] == 1: # suhan's scenario fig = px.line(df, x=df.index, y=df.columns, markers=True) fig.update_layout(xaxis_title="Loop Round", yaxis_title=None) st.plotly_chart(fig) else: metrics_window(df, 1, 4, height=300, colors=["red", "blue", "orange", "green"]) elif state.log_type == "Model from Paper" and len(state.msgs[state.lround]["d.evolving code"]) > 0: with st.container(border=True): st.subheader("Summary📊", divider="rainbow", anchor="_summary") # pass ws: list[FactorFBWorkspace | ModelFBWorkspace] = state.msgs[state.lround]["d.evolving code"][-1].content # All Tasks tab_names = [ w.target_task.factor_name if isinstance(w.target_task, FactorTask) else w.target_task.name for w in ws ] for j in range(len(ws)): if state.msgs[state.lround]["d.evolving feedback"][-1].content[j].final_decision: tab_names[j] += "✔️" else: tab_names[j] += "❌" wtabs = st.tabs(tab_names) for j, w in enumerate(ws): with wtabs[j]: # Evolving Code for k, v in w.code_dict.items(): with st.expander(f":green[`{k}`]", expanded=False): st.code(v, language="python") # Evolving Feedback evolving_feedback_window(state.msgs[state.lround]["d.evolving feedback"][-1].content[j]) def tabs_hint(): st.markdown( "

You can navigate through the tabs using ⬅️ ➡️ or by holding Shift and scrolling with the mouse wheel🖱️.

", unsafe_allow_html=True, ) # TODO: when tab names are too long, some tabs are not shown def tasks_window(tasks: list[FactorTask | ModelTask]): if isinstance(tasks[0], FactorTask): st.markdown("**Factor Tasks🚩**") tnames = [f.factor_name for f in tasks] if sum(len(tn) for tn in tnames) > 100: tabs_hint() tabs = st.tabs(tnames) for i, ft in enumerate(tasks): with tabs[i]: # st.markdown(f"**Factor Name**: {ft.factor_name}") st.markdown(f"**Description**: {ft.factor_description}") st.latex("Formulation") st.latex(f"{ft.factor_formulation}") mks = "| Variable | Description |\n| --- | --- |\n" for v, d in ft.variables.items(): mks += f"| ${v}$ | {d} |\n" st.markdown(mks) elif isinstance(tasks[0], ModelTask): st.markdown("**Model Tasks🚩**") tnames = [m.name for m in tasks] if sum(len(tn) for tn in tnames) > 100: tabs_hint() tabs = st.tabs(tnames) for i, mt in enumerate(tasks): with tabs[i]: # st.markdown(f"**Model Name**: {mt.name}") st.markdown(f"**Model Type**: {mt.model_type}") st.markdown(f"**Description**: {mt.description}") st.latex("Formulation") st.latex(f"{mt.formulation}") mks = "| Variable | Description |\n| --- | --- |\n" for v, d in mt.variables.items(): mks += f"| ${v}$ | {d} |\n" st.markdown(mks) # Config Sidebar with st.sidebar: st.markdown( """ # RD-Agent🤖 ## [Scenario Description](#_scenario) ## [Summary](#_summary) - [**Hypotheses**](#_hypotheses) - [**Metrics**](#_metrics) ## [RD-Loops](#_rdloops) - [**Research**](#_research) - [**Development**](#_development) - [**Feedback**](#_feedback) """ ) st.selectbox( ":green[**Scenario**]", ["Qlib Model", "Data Mining", "Qlib Factor", "Model from Paper"], key="log_type" ) with st.popover(":orange[**Config⚙️**]"): with st.container(border=True): st.markdown(":blue[**log path**]") if main_log_path: if st.toggle("Manual Input"): st.text_input("log path", key="log_path", on_change=refresh) else: folders = [ folder.relative_to(main_log_path) for folder in main_log_path.iterdir() if folder.is_dir() ] st.selectbox(f"Select from `{main_log_path}`", folders, key="log_path", on_change=refresh) else: st.text_input("log path", key="log_path", on_change=refresh) with st.container(border=True): st.markdown(":blue[**excluded configs**]") st.multiselect("excluded log tags", ["llm_messages"], ["llm_messages"], key="excluded_tags") st.multiselect("excluded log types", ["str", "dict", "list"], ["str"], key="excluded_types") if st.button("All Loops"): if not state.fs: refresh() get_msgs_until(lambda m: False) if st.button("Next Loop"): if not state.fs: refresh() get_msgs_until(lambda m: "ef.feedback" in m.tag) if st.button("One Evolving"): if not state.fs: refresh() get_msgs_until(lambda m: "d.evolving feedback" in m.tag) if st.button("refresh logs", help="clear all log messages in cache"): refresh() debug = st.toggle("debug", value=False) if debug: if st.button("Single Step Run"): if not state.fs: refresh() get_msgs_until() # Debug Info Window if debug: with st.expander(":red[**Debug Info**]", expanded=True): dcol1, dcol2 = st.columns([1, 3]) with dcol1: st.markdown( f"**trace type**: {state.log_type}\n\n" f"**log path**: {state.log_path}\n\n" f"**excluded tags**: {state.excluded_tags}\n\n" f"**excluded types**: {state.excluded_types}\n\n" f":blue[**message id**]: {sum(sum(len(tmsgs) for tmsgs in rmsgs.values()) for rmsgs in state.msgs.values())}\n\n" f":blue[**round**]: {state.lround}\n\n" f":blue[**evolving round**]: {state.erounds[state.lround]}\n\n" ) with dcol2: if state.last_msg: st.write(state.last_msg) if isinstance(state.last_msg.content, list): st.write(state.last_msg.content[0]) elif not isinstance(state.last_msg.content, str): st.write(state.last_msg.content.__dict__) # Main Window header_c1, header_c3 = st.columns([1, 6], vertical_alignment="center") with st.container(): with header_c1: st.image("https://img-prod-cms-rt-microsoft-com.akamaized.net/cms/api/am/imageFileData/RE1Mu3b?ver=5c31") with header_c3: st.markdown( """

RD-Agent:
LLM-based autonomous evolving agents for industrial data-driven R&D

""", unsafe_allow_html=True, ) # Project Info with st.container(): image_c, scen_c = st.columns([3, 3], vertical_alignment="center") with image_c: st.image("./docs/_static/flow.png") with scen_c: st.header("Scenario Description📖", divider="violet", anchor="_scenario") # TODO: other scenarios if state.log_type == "Qlib Model": st.markdown(QlibModelScenario().rich_style_description) elif state.log_type == "Data Mining": st.markdown(DMModelScenario().rich_style_description) elif state.log_type == "Qlib Factor": st.markdown(QlibFactorScenario().rich_style_description) elif state.log_type == "Model from Paper": st.markdown(GeneralModelScenario().rich_style_description) # Summary Window summary_window() # R&D Loops Window if state.log_type in ["Qlib Model", "Data Mining", "Qlib Factor"]: st.header("R&D Loops♾️", divider="rainbow", anchor="_rdloops") if state.log_type in ["Qlib Model", "Data Mining", "Qlib Factor"]: if len(state.msgs) > 1: r_options = list(state.msgs.keys()) if 0 in r_options: r_options.remove(0) round = st_btn_select(options=r_options, index=state.lround - 1) else: round = 1 else: round = 1 def research_window(): with st.container(border=True): title = "Research🔍" if state.log_type in ["Qlib Model", "Data Mining", "Qlib Factor"] else "Research🔍 (reader)" st.subheader(title, divider="blue", anchor="_research") if state.log_type in ["Qlib Model", "Data Mining", "Qlib Factor"]: # pdf image if pim := state.msgs[round]["r.extract_factors_and_implement.load_pdf_screenshot"]: for i in range(min(2, len(pim))): st.image(pim[i].content, use_column_width=True) # Hypothesis if hg := state.msgs[round]["r.hypothesis generation"]: st.markdown("**Hypothesis💡**") # 🧠 h: Hypothesis = hg[0].content st.markdown( f""" - **Hypothesis**: {h.hypothesis} - **Reason**: {h.reason}""" ) if eg := state.msgs[round]["r.experiment generation"]: tasks_window(eg[0].content) elif state.log_type == "Model from Paper": # pdf image c1, c2 = st.columns([2, 3]) with c1: if pim := state.msgs[round]["r.pdf_image"]: for i in range(len(pim)): st.image(pim[i].content, use_column_width=True) # loaded model exp with c2: if mem := state.msgs[round]["d.load_experiment"]: me: QlibModelExperiment = mem[0].content tasks_window(me.sub_tasks) def feedback_window(): if state.log_type in ["Qlib Model", "Data Mining", "Qlib Factor"]: with st.container(border=True): st.subheader("Feedback📝", divider="orange", anchor="_feedback") if fbr := state.msgs[round]["ef.Quantitative Backtesting Chart"]: st.markdown("**Returns📈**") fig = report_figure(fbr[0].content) st.plotly_chart(fig) if fb := state.msgs[round]["ef.feedback"]: st.markdown("**Hypothesis Feedback🔍**") h: HypothesisFeedback = fb[0].content st.markdown( f""" - **Observations**: {h.observations} - **Hypothesis Evaluation**: {h.hypothesis_evaluation} - **New Hypothesis**: {h.new_hypothesis} - **Decision**: {h.decision} - **Reason**: {h.reason}""" ) if state.log_type in ["Qlib Model", "Data Mining", "Qlib Factor"]: rf_c, d_c = st.columns([2, 2]) elif state.log_type == "Model from Paper": rf_c = st.container() d_c = st.container() with rf_c: research_window() feedback_window() # Development Window (Evolving) with d_c.container(border=True): title = ( "Development🛠️" if state.log_type in ["Qlib Model", "Data Mining", "Qlib Factor"] else "Development🛠️ (evolving coder)" ) st.subheader(title, divider="green", anchor="_development") # Evolving Status if state.erounds[round] > 0: st.markdown("**☑️ Evolving Status**") es = state.e_decisions[round] e_status_mks = "".join(f"| {ei} " for ei in range(1, state.erounds[round] + 1)) + "|\n" e_status_mks += "|--" * state.erounds[round] + "|\n" for ei, estatus in es.items(): if not estatus: estatus = (0, 0) e_status_mks += "| " + "✔️
" * estatus[0] + "❌
" * estatus[1] + " " e_status_mks += "|\n" st.markdown(e_status_mks, unsafe_allow_html=True) # Evolving Tabs if state.erounds[round] > 0: if state.erounds[round] > 1: st.markdown("**🔄️Evolving Rounds**") evolving_round = st_btn_select( options=range(1, state.erounds[round] + 1), index=state.erounds[round] - 1, key="show_eround" ) else: evolving_round = 1 ws: list[FactorFBWorkspace | ModelFBWorkspace] = state.msgs[round]["d.evolving code"][ evolving_round - 1 ].content # All Tasks tab_names = [ w.target_task.factor_name if isinstance(w.target_task, FactorTask) else w.target_task.name for w in ws ] if len(state.msgs[round]["d.evolving feedback"]) >= evolving_round: for j in range(len(ws)): if state.msgs[round]["d.evolving feedback"][evolving_round - 1].content[j].final_decision: tab_names[j] += "✔️" else: tab_names[j] += "❌" if sum(len(tn) for tn in tab_names) > 100: tabs_hint() wtabs = st.tabs(tab_names) for j, w in enumerate(ws): with wtabs[j]: # Evolving Code for k, v in w.code_dict.items(): with st.expander(f":green[`{k}`]", expanded=True): st.code(v, language="python") # Evolving Feedback if len(state.msgs[round]["d.evolving feedback"]) >= evolving_round: evolving_feedback_window(state.msgs[round]["d.evolving feedback"][evolving_round - 1].content[j]) with st.container(border=True): st.subheader("Disclaimer", divider="gray") st.markdown( "This content is AI-generated and may not be fully accurate or up-to-date; please verify with a professional for critical matters." )