TradingAgents/tradingagents/graph/parallel_analysts.py

232 lines
8.5 KiB
Python

"""Parallel execution nodes for TradingAgents.
Provides parallel wrappers for:
- Analyst phase (Market, Social, News, Fundamentals)
- Research debate phase (Bull + Bear)
- Risk debate phase (Aggressive + Conservative + Neutral)
"""
import asyncio
import logging
from concurrent.futures import ThreadPoolExecutor
from langchain_core.messages import HumanMessage, RemoveMessage
logger = logging.getLogger(__name__)
def create_parallel_analyst_node(analyst_fns, tool_nodes, selected_analysts):
"""Create a single LangGraph node that runs all analysts in parallel.
Each analyst gets its own isolated message state and runs its complete
tool-calling loop independently. Results are merged at the end.
Args:
analyst_fns: dict mapping analyst type (e.g. "market") to node function
tool_nodes: dict mapping analyst type to ToolNode instance
selected_analysts: list of analyst types to run
"""
async def parallel_analysts_node(state):
"""Run all analysts concurrently and merge their reports."""
async def run_single(analyst_type):
"""Run one analyst through its complete tool-calling loop."""
fn = analyst_fns[analyst_type]
tn = tool_nodes[analyst_type]
# Each analyst gets its own isolated message state
local_state = {
"messages": list(state["messages"]),
"trade_date": state["trade_date"],
"company_of_interest": state["company_of_interest"],
}
result = {}
for _ in range(10): # safety limit on tool rounds
result = await asyncio.to_thread(fn, local_state)
ai_msg = result["messages"][0]
local_state["messages"] = local_state["messages"] + [ai_msg]
if not ai_msg.tool_calls:
break
# Process tool calls
tool_result = await asyncio.to_thread(tn.invoke, local_state)
local_state["messages"] = (
local_state["messages"] + tool_result["messages"]
)
# Return only report fields (not messages)
return {k: v for k, v in result.items() if k != "messages"}
# Run all analysts concurrently
tasks = [run_single(at) for at in selected_analysts if at in analyst_fns]
results = await asyncio.gather(*tasks)
# Merge all report fields
merged = {}
for r in results:
merged.update(r)
# Clear messages and add placeholder (same as Msg Clear nodes)
messages = state.get("messages", [])
removal_ops = [
RemoveMessage(id=m.id)
for m in messages
if hasattr(m, "id") and m.id
]
merged["messages"] = removal_ops + [HumanMessage(content="Continue")]
return merged
return parallel_analysts_node
def _snapshot_research_state(state):
"""Extract research-relevant fields into a plain dict."""
return {
"investment_debate_state": dict(state.get("investment_debate_state", {})),
"market_report": state.get("market_report", ""),
"sentiment_report": state.get("sentiment_report", ""),
"news_report": state.get("news_report", ""),
"fundamentals_report": state.get("fundamentals_report", ""),
}
def _snapshot_risk_state(state):
"""Extract risk-relevant fields into a plain dict."""
return {
"risk_debate_state": dict(state.get("risk_debate_state", {})),
"market_report": state.get("market_report", ""),
"sentiment_report": state.get("sentiment_report", ""),
"news_report": state.get("news_report", ""),
"fundamentals_report": state.get("fundamentals_report", ""),
"trader_investment_plan": state.get("trader_investment_plan", ""),
}
def create_parallel_research_node(bull_fn, bear_fn):
"""Create a node that runs Bull and Bear researchers in parallel.
Uses a sync function with ThreadPoolExecutor.submit() to avoid any
asyncio event-loop interaction issues. LangGraph handles running sync
nodes in its own thread, and from there we spawn our own pool.
"""
def parallel_research_node(state):
import time
state_snap = _snapshot_research_state(state)
t0 = time.time()
def run_bull():
print(f"[PARALLEL] Bull starting at +{time.time()-t0:.1f}s", flush=True)
result = bull_fn(state_snap)
print(f"[PARALLEL] Bull done at +{time.time()-t0:.1f}s", flush=True)
return result
def run_bear():
print(f"[PARALLEL] Bear starting at +{time.time()-t0:.1f}s", flush=True)
result = bear_fn(state_snap)
print(f"[PARALLEL] Bear done at +{time.time()-t0:.1f}s", flush=True)
return result
with ThreadPoolExecutor(max_workers=2) as pool:
bull_future = pool.submit(run_bull)
bear_future = pool.submit(run_bear)
bull_result = bull_future.result()
bear_result = bear_future.result()
print(f"[PARALLEL] Research total: {time.time()-t0:.1f}s", flush=True)
bull_debate = bull_result["investment_debate_state"]
bear_debate = bear_result["investment_debate_state"]
merged_debate = {
"bull_history": bull_debate.get("bull_history", ""),
"bear_history": bear_debate.get("bear_history", ""),
"history": bull_debate.get("bull_history", "")
+ "\n"
+ bear_debate.get("bear_history", ""),
"current_response": bear_debate.get("current_response", ""),
"judge_decision": "",
"count": 2,
}
return {"investment_debate_state": merged_debate}
return parallel_research_node
def create_parallel_risk_node(aggressive_fn, conservative_fn, neutral_fn):
"""Create a node that runs all 3 risk analysts in parallel.
Uses a sync function with ThreadPoolExecutor.submit() to avoid any
asyncio event-loop interaction issues. LangGraph handles running sync
nodes in its own thread, and from there we spawn our own pool.
"""
def parallel_risk_node(state):
import time
state_snap = _snapshot_risk_state(state)
t0 = time.time()
def run_agg():
print(f"[PARALLEL] Aggressive starting at +{time.time()-t0:.1f}s", flush=True)
result = aggressive_fn(state_snap)
print(f"[PARALLEL] Aggressive done at +{time.time()-t0:.1f}s", flush=True)
return result
def run_con():
print(f"[PARALLEL] Conservative starting at +{time.time()-t0:.1f}s", flush=True)
result = conservative_fn(state_snap)
print(f"[PARALLEL] Conservative done at +{time.time()-t0:.1f}s", flush=True)
return result
def run_neu():
print(f"[PARALLEL] Neutral starting at +{time.time()-t0:.1f}s", flush=True)
result = neutral_fn(state_snap)
print(f"[PARALLEL] Neutral done at +{time.time()-t0:.1f}s", flush=True)
return result
with ThreadPoolExecutor(max_workers=3) as pool:
agg_future = pool.submit(run_agg)
con_future = pool.submit(run_con)
neu_future = pool.submit(run_neu)
agg_result = agg_future.result()
con_result = con_future.result()
neu_result = neu_future.result()
logger.info("Parallel risk total: %.1fs", time.time() - t0)
agg_debate = agg_result["risk_debate_state"]
con_debate = con_result["risk_debate_state"]
neu_debate = neu_result["risk_debate_state"]
merged_debate = {
"aggressive_history": agg_debate.get("aggressive_history", ""),
"conservative_history": con_debate.get("conservative_history", ""),
"neutral_history": neu_debate.get("neutral_history", ""),
"history": agg_debate.get("aggressive_history", "")
+ "\n"
+ con_debate.get("conservative_history", "")
+ "\n"
+ neu_debate.get("neutral_history", ""),
"latest_speaker": "Neutral",
"current_aggressive_response": agg_debate.get(
"current_aggressive_response", ""
),
"current_conservative_response": con_debate.get(
"current_conservative_response", ""
),
"current_neutral_response": neu_debate.get(
"current_neutral_response", ""
),
"judge_decision": "",
"count": 3,
}
return {"risk_debate_state": merged_debate}
return parallel_risk_node