103 lines
3.5 KiB
Python
103 lines
3.5 KiB
Python
# -*- coding: utf-8 -*-
|
|
import time
|
|
import json
|
|
import logging
|
|
import random
|
|
from tenacity import (
|
|
retry,
|
|
stop_after_attempt,
|
|
wait_exponential,
|
|
retry_if_exception_type,
|
|
before_sleep_log
|
|
)
|
|
from anthropic._exceptions import OverloadedError
|
|
from tradingagents.agents.utils.output_filter import fix_common_llm_errors, validate_and_warn
|
|
from tradingagents.agents.utils.prompts import get_research_manager_prompt
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def create_research_manager(llm, memory, language: str = "zh-TW"):
|
|
"""
|
|
建立一個研究管理員（裁判）節點。
|
|
|
|
Args:
|
|
llm: 用於生成決策和計畫的語言模型。
|
|
memory: 儲存過去情況和反思的記憶體物件。
|
|
language: 報告語言 ('en' 或 'zh-TW')
|
|
|
|
Returns:
|
|
function: 一個代表研究管理員節點的函式。
|
|
"""
|
|
|
|
def research_manager_node(state) -> dict:
|
|
"""研究管理員節點的執行函式。"""
|
|
investment_debate_state = state["investment_debate_state"]
|
|
history = investment_debate_state.get("history", "")
|
|
|
|
market_research_report = state["market_report"]
|
|
sentiment_report = state["sentiment_report"]
|
|
news_report = state["news_report"]
|
|
fundamentals_report = state["fundamentals_report"]
|
|
|
|
curr_situation = f"{market_research_report}\n\n{sentiment_report}\n\n{news_report}\n\n{fundamentals_report}"
|
|
|
|
past_memories = memory.get_memories(curr_situation, n_matches=2)
|
|
past_memory_str = ""
|
|
for i, rec in enumerate(past_memories, 1):
|
|
recommendation = rec["recommendation"]
|
|
past_memory_str += recommendation + "\n\n"
|
|
|
|
# Get language-specific prompt
|
|
base_prompt = get_research_manager_prompt(language)
|
|
|
|
if language == "en":
|
|
prompt = f"""{base_prompt}
|
|
|
|
【Available Information】
|
|
- Past Reflections: "{past_memory_str}"
|
|
- Debate History: {history}
|
|
|
|
Please provide your investment decision report."""
|
|
else:
|
|
prompt = f"""{base_prompt}
|
|
|
|
【可用資訊】
|
|
- 過去反思："{past_memory_str}"
|
|
- 辯論歷史：{history}
|
|
|
|
請提供您的投資決策報告。"""
|
|
|
|
@retry(
|
|
retry=retry_if_exception_type(OverloadedError),
|
|
wait=wait_exponential(multiplier=1, min=2, max=60),
|
|
stop=stop_after_attempt(5),
|
|
before_sleep=before_sleep_log(logger, logging.WARNING)
|
|
)
|
|
def invoke_llm_with_retry(llm_instance, prompt_text):
|
|
jitter = random.uniform(0, 0.5)
|
|
if jitter > 0:
|
|
time.sleep(jitter)
|
|
logger.info("正在調用 Research Manager LLM...")
|
|
return llm_instance.invoke(prompt_text)
|
|
|
|
response = invoke_llm_with_retry(llm, prompt)
|
|
|
|
response.content = fix_common_llm_errors(response.content)
|
|
validate_and_warn(response.content, "Research_Manager")
|
|
|
|
new_investment_debate_state = {
|
|
"judge_decision": response.content,
|
|
"history": investment_debate_state.get("history", ""),
|
|
"bear_history": investment_debate_state.get("bear_history", ""),
|
|
"bull_history": investment_debate_state.get("bull_history", ""),
|
|
"current_response": response.content,
|
|
"count": investment_debate_state["count"],
|
|
}
|
|
|
|
return {
|
|
"investment_debate_state": new_investment_debate_state,
|
|
"investment_plan": response.content,
|
|
}
|
|
|
|
return research_manager_node |