SII-leiyuandClaude
0254e2c01a
- Add trade monitoring and whale detection system - Add LLM-powered trade analysis - Add market data fetching from Polymarket API - Include example environment configuration - Add automated report generation 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
300 lines
10 KiB
Python
300 lines
10 KiB
Python
"""LLM analyzer service - analyzes whale trades using AI."""
|
|
import json
|
|
import logging
|
|
import os
|
|
import re
|
|
from datetime import datetime
|
|
from typing import Optional
|
|
|
|
from google import genai
|
|
from google.genai import types
|
|
|
|
from src.config import get_settings
|
|
from src.models.trade import WhaleTrade
|
|
from src.models.decision import LLMDecision, TradeRecommendation, TradeAction, TraderCredibility
|
|
from src.services.anomaly_detector import AnomalyDetector
|
|
from src.prompts.whale_analyzer import WhaleAnalyzerPrompts
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class LLMAnalyzer:
|
|
"""
|
|
Analyzes whale trades using LLM (Google Gemini models).
|
|
|
|
Combines trade context with superforecaster methodology to generate
|
|
comprehensive analysis reports with trading recommendations.
|
|
"""
|
|
|
|
def __init__(self):
|
|
self.settings = get_settings()
|
|
|
|
# Configure Gemini API using new client SDK
|
|
os.environ["GOOGLE_API_KEY"] = self.settings.gemini_api_key
|
|
self.client = genai.Client()
|
|
|
|
self.anomaly_detector = AnomalyDetector()
|
|
self.prompts = WhaleAnalyzerPrompts()
|
|
|
|
def _extract_json_from_response(self, response: str) -> Optional[dict]:
|
|
"""
|
|
Extract JSON from LLM response.
|
|
|
|
Args:
|
|
response: The LLM response text
|
|
|
|
Returns:
|
|
Parsed JSON dict or None
|
|
"""
|
|
# Try to find JSON in code blocks
|
|
json_pattern = r"```(?:json)?\s*([\s\S]*?)```"
|
|
matches = re.findall(json_pattern, response)
|
|
|
|
for match in matches:
|
|
try:
|
|
return json.loads(match.strip())
|
|
except json.JSONDecodeError:
|
|
continue
|
|
|
|
# Try to find raw JSON
|
|
try:
|
|
# Find JSON-like content
|
|
start = response.find("{")
|
|
end = response.rfind("}") + 1
|
|
if start >= 0 and end > start:
|
|
return json.loads(response[start:end])
|
|
except json.JSONDecodeError:
|
|
pass
|
|
|
|
return None
|
|
|
|
def _parse_recommendation(self, json_data: dict) -> TradeRecommendation:
|
|
"""
|
|
Parse JSON data into TradeRecommendation.
|
|
|
|
Args:
|
|
json_data: Parsed JSON from LLM
|
|
|
|
Returns:
|
|
TradeRecommendation object
|
|
"""
|
|
action_str = json_data.get("action", "HOLD").upper()
|
|
try:
|
|
action = TradeAction(action_str)
|
|
except ValueError:
|
|
action = TradeAction.HOLD
|
|
|
|
confidence = float(json_data.get("confidence", 0.0))
|
|
# Clamp confidence to valid range
|
|
confidence = max(0.0, min(1.0, confidence))
|
|
|
|
suggested_price = json_data.get("suggested_price")
|
|
if suggested_price is not None:
|
|
suggested_price = float(suggested_price)
|
|
|
|
suggested_size = float(json_data.get("suggested_size_percent", 0.1))
|
|
suggested_size = max(0.0, min(1.0, suggested_size))
|
|
|
|
# Parse insider trading assessment fields
|
|
insider_likelihood = float(json_data.get("insider_trading_likelihood", 0.0))
|
|
insider_likelihood = max(0.0, min(1.0, insider_likelihood))
|
|
|
|
credibility_str = json_data.get("trader_credibility", "UNKNOWN").upper()
|
|
try:
|
|
trader_credibility = TraderCredibility(credibility_str)
|
|
except ValueError:
|
|
trader_credibility = TraderCredibility.UNKNOWN
|
|
|
|
insider_evidence = str(json_data.get("insider_evidence", ""))
|
|
|
|
return TradeRecommendation(
|
|
action=action,
|
|
outcome=str(json_data.get("outcome", "")),
|
|
confidence=confidence,
|
|
suggested_price=suggested_price,
|
|
suggested_size_percent=suggested_size,
|
|
reasoning=str(json_data.get("reasoning", "")),
|
|
insider_trading_likelihood=insider_likelihood,
|
|
trader_credibility=trader_credibility,
|
|
insider_evidence=insider_evidence,
|
|
)
|
|
|
|
async def analyze_whale_trade(self, whale_trade: WhaleTrade) -> LLMDecision:
|
|
"""
|
|
Analyze a whale trade using LLM.
|
|
|
|
Args:
|
|
whale_trade: The whale trade to analyze
|
|
|
|
Returns:
|
|
LLMDecision with analysis and recommendation
|
|
"""
|
|
# Format trade context for LLM
|
|
trade_context = self.anomaly_detector.format_for_llm(whale_trade)
|
|
|
|
# Build prompt (Gemini uses single prompt with system instruction)
|
|
system_prompt = self.prompts.system_prompt()
|
|
user_prompt = self.prompts.analyze_whale_trade(trade_context)
|
|
full_prompt = f"{system_prompt}\n\n---\n\n{user_prompt}"
|
|
|
|
try:
|
|
# Call Gemini API with Google Search tool enabled
|
|
response = self.client.models.generate_content(
|
|
model=self.settings.llm_model,
|
|
contents=full_prompt,
|
|
config=types.GenerateContentConfig(
|
|
tools=[types.Tool(google_search=types.GoogleSearch())],
|
|
),
|
|
)
|
|
|
|
analysis_text = response.text
|
|
logger.debug(f"LLM response: {analysis_text[:500]}...")
|
|
|
|
# Extract JSON from response
|
|
json_data = self._extract_json_from_response(analysis_text)
|
|
|
|
if json_data:
|
|
recommendation = self._parse_recommendation(json_data)
|
|
else:
|
|
# Default to HOLD if we can't parse the response
|
|
logger.warning("Could not parse LLM response as JSON, defaulting to HOLD")
|
|
recommendation = TradeRecommendation(
|
|
action=TradeAction.HOLD,
|
|
outcome="",
|
|
confidence=0.0,
|
|
reasoning="Failed to parse LLM response",
|
|
)
|
|
|
|
return LLMDecision(
|
|
whale_trade_id=whale_trade.id,
|
|
market_id=whale_trade.market_id,
|
|
analysis=analysis_text,
|
|
recommendation=recommendation,
|
|
)
|
|
|
|
except Exception as e:
|
|
logger.error(f"Error calling LLM: {e}")
|
|
# Return a safe default decision
|
|
return LLMDecision(
|
|
whale_trade_id=whale_trade.id,
|
|
market_id=whale_trade.market_id,
|
|
analysis=f"Error during analysis: {str(e)}",
|
|
recommendation=TradeRecommendation(
|
|
action=TradeAction.HOLD,
|
|
outcome="",
|
|
confidence=0.0,
|
|
reasoning=f"Analysis failed: {str(e)}",
|
|
),
|
|
)
|
|
|
|
def format_full_report(self, whale_trade: WhaleTrade, decision: LLMDecision) -> str:
|
|
"""
|
|
Format a complete analysis report with trade info, analysis, and decision.
|
|
|
|
Args:
|
|
whale_trade: The whale trade
|
|
decision: The LLM decision
|
|
|
|
Returns:
|
|
Formatted report string
|
|
"""
|
|
trade = whale_trade.trade
|
|
rec = decision.recommendation
|
|
|
|
# Format outcome prices
|
|
prices_str = ""
|
|
if whale_trade.market_outcomes and whale_trade.market_outcome_prices:
|
|
prices_str = " | ".join([
|
|
f"{o}: {p:.1%}"
|
|
for o, p in zip(whale_trade.market_outcomes, whale_trade.market_outcome_prices)
|
|
])
|
|
|
|
# Action emoji and color indicator
|
|
action_indicator = {
|
|
TradeAction.BUY: "🟢 BUY",
|
|
TradeAction.SELL: "🔴 SELL",
|
|
TradeAction.HOLD: "⚪ HOLD",
|
|
}
|
|
|
|
# Insider trading likelihood indicator
|
|
insider_likelihood = rec.insider_trading_likelihood
|
|
if insider_likelihood >= 0.7:
|
|
insider_indicator = f"🔴 高度可疑 ({insider_likelihood:.0%})"
|
|
elif insider_likelihood >= 0.4:
|
|
insider_indicator = f"🟡 中等可能 ({insider_likelihood:.0%})"
|
|
else:
|
|
insider_indicator = f"🟢 普通交易 ({insider_likelihood:.0%})"
|
|
|
|
# Trader credibility indicator
|
|
credibility_indicators = {
|
|
TraderCredibility.HIGH: "🏆 高可信度 (前100名)",
|
|
TraderCredibility.MEDIUM: "⭐ 中等可信度 (100-500名)",
|
|
TraderCredibility.LOW: "📉 低可信度 (500名+)",
|
|
TraderCredibility.UNKNOWN: "❓ 未知 (未上榜)",
|
|
}
|
|
credibility_str = credibility_indicators.get(rec.trader_credibility, "❓ 未知")
|
|
|
|
# Trader ranking info
|
|
trader_ranking_str = ""
|
|
if whale_trade.trader_ranking:
|
|
tr = whale_trade.trader_ranking
|
|
rank_str = f"#{tr.rank}" if tr.rank else "未上榜"
|
|
pnl_str = f"${tr.pnl:,.2f}" if tr.pnl else "N/A"
|
|
trader_ranking_str = f"| **交易者排名** | {rank_str} (PnL: {pnl_str}) |"
|
|
|
|
report = f"""
|
|
{'='*70}
|
|
# 🐋 鲸鱼交易分析报告
|
|
{'='*70}
|
|
|
|
**生成时间**: {datetime.utcnow().strftime('%Y-%m-%d %H:%M:%S')} UTC
|
|
|
|
## 交易摘要
|
|
|
|
| 项目 | 详情 |
|
|
|------|------|
|
|
| **市场** | {whale_trade.market_question} |
|
|
| **交易金额** | ${trade.usdc_size:,.2f} USDC |
|
|
| **交易方向** | {trade.side} |
|
|
| **交易价格** | {trade.price:.4f} ({trade.price:.1%}) |
|
|
| **交易结果** | {trade.outcome} |
|
|
| **当前赔率** | {prices_str} |
|
|
| **交易时间** | {datetime.fromtimestamp(trade.timestamp).strftime('%Y-%m-%d %H:%M:%S') if trade.timestamp else 'N/A'} |
|
|
{trader_ranking_str}
|
|
|
|
{'='*70}
|
|
|
|
{decision.analysis}
|
|
|
|
{'='*70}
|
|
## 🔍 内幕交易评估
|
|
{'='*70}
|
|
|
|
| 项目 | 评估 |
|
|
|------|------|
|
|
| **内幕交易可能性** | {insider_indicator} |
|
|
| **交易者可信度** | {credibility_str} |
|
|
|
|
**关键证据**: {rec.insider_evidence or '无明确证据'}
|
|
|
|
{'='*70}
|
|
## 📊 决策摘要
|
|
{'='*70}
|
|
|
|
| 项目 | 建议 |
|
|
|------|------|
|
|
| **操作建议** | {action_indicator.get(rec.action, '⚪ HOLD')} |
|
|
| **目标结果** | {rec.outcome or 'N/A'} |
|
|
| **信心程度** | {rec.confidence:.1%} |
|
|
| **建议仓位** | {rec.suggested_size_percent:.1%} |
|
|
| **建议价格** | {f'{rec.suggested_price:.4f}' if rec.suggested_price else 'Market'} |
|
|
|
|
**决策理由**: {rec.reasoning}
|
|
|
|
{'='*70}
|
|
⚠️ 免责声明：本报告由AI生成，仅供参考，不构成投资建议。
|
|
预测市场具有高风险，请基于自身判断谨慎决策。
|
|
{'='*70}
|
|
"""
|
|
return report
|