146 lines
5.3 KiB
Python
146 lines
5.3 KiB
Python
import os
|
|
import json
|
|
import numpy as np
|
|
from modules.services.ollama import OllamaManager
|
|
from modules.analysis.technical import TechnicalAnalyzer
|
|
from modules.analysis.deep_learning import PricePredictor
|
|
|
|
# [최적화] 워커 프로세스별 전역 변수 (LSTM 모델 캐싱)
|
|
_lstm_predictor = None
|
|
|
|
def get_predictor():
|
|
"""워커 프로세스 내에서 PricePredictor 인스턴스를 싱글톤으로 관리"""
|
|
global _lstm_predictor
|
|
if _lstm_predictor is None:
|
|
print(f"🧩 [Worker {os.getpid()}] Initializing LSTM Predictor...")
|
|
_lstm_predictor = PricePredictor()
|
|
return _lstm_predictor
|
|
|
|
def analyze_stock_process(ticker, prices, news_items):
|
|
"""
|
|
[CPU Intensive] 기술적 분석 및 AI 판단을 수행하는 함수
|
|
(ProcessPoolExecutor에서 실행됨)
|
|
"""
|
|
print(f"⚙️ [Bot Process] Analyzing {ticker} ({len(prices)} candles)...")
|
|
|
|
# 1. 기술적 지표 계산
|
|
current_price = prices[-1] if prices else 0
|
|
# [수정] 변동성, 거래량 비율 반환 (거래량 데이터가 없으면 None 전달)
|
|
tech_score, rsi, volatility, vol_ratio = TechnicalAnalyzer.get_technical_score(current_price, prices, volume_history=None)
|
|
|
|
# 2. LSTM 주가 예측
|
|
# [최적화] 전역 캐시된 Predictor 사용
|
|
lstm_predictor = get_predictor()
|
|
if lstm_predictor:
|
|
lstm_predictor.training_status['current_ticker'] = ticker
|
|
pred_result = lstm_predictor.train_and_predict(prices)
|
|
|
|
lstm_score = 0.5 # 중립
|
|
ai_confidence = 0.5
|
|
ai_loss = 1.0
|
|
|
|
if pred_result:
|
|
ai_confidence = pred_result.get('confidence', 0.5)
|
|
ai_loss = pred_result.get('loss', 1.0)
|
|
|
|
# 상승/하락 예측에 따라 점수 조정 (신뢰도 반영)
|
|
# 최대 5% 변동폭까지 반영
|
|
change_magnitude = min(abs(pred_result['change_rate']), 5.0) / 5.0
|
|
|
|
if pred_result['trend'] == 'UP':
|
|
# 상승 예측 시: 기본 0.5 + (강도 * 신뢰도 * 0.4) -> 최대 0.9
|
|
lstm_score = 0.5 + (change_magnitude * ai_confidence * 0.4)
|
|
else:
|
|
# 하락 예측 시: 기본 0.5 - (강도 * 신뢰도 * 0.4) -> 최소 0.1
|
|
lstm_score = 0.5 - (change_magnitude * ai_confidence * 0.4)
|
|
|
|
lstm_score = max(0.0, min(1.0, lstm_score))
|
|
|
|
# 3. AI 뉴스 분석
|
|
ollama = OllamaManager()
|
|
prompt = f"""
|
|
[System Instruction]
|
|
1. Role: You are a Expert Quant Trader with 20 years of experience.
|
|
2. Market Data:
|
|
- Technical Score: {tech_score:.2f} (RSI: {rsi:.2f})
|
|
- AI Prediction: {pred_result['predicted']:.0f} KRW ({pred_result['change_rate']}%)
|
|
- AI Confidence: {ai_confidence:.2f} (Loss: {ai_loss:.4f})
|
|
3. Strategy:
|
|
- If AI Confidence > 0.8 and Trend is UP -> Strong BUY signal.
|
|
- If Tech Score > 0.7 -> BUY signal.
|
|
- If Trend is DOWN -> SELL/AVOID.
|
|
4. Task: Analyze the news and combine with market data to decide sentiment.
|
|
|
|
News Data: {json.dumps(news_items, ensure_ascii=False)}
|
|
|
|
Response (JSON):
|
|
{{
|
|
"sentiment_score": 0.8,
|
|
"reason": "AI confidence is high and news supports the uptrend."
|
|
}}
|
|
"""
|
|
ai_resp = ollama.request_inference(prompt)
|
|
sentiment_score = 0.5
|
|
try:
|
|
data = json.loads(ai_resp)
|
|
sentiment_score = float(data.get("sentiment_score", 0.5))
|
|
except:
|
|
pass
|
|
|
|
# 4. 통합 점수 (동적 가중치)
|
|
# AI 신뢰도가 높으면 AI 비중을 대폭 상향
|
|
if ai_confidence >= 0.85:
|
|
w_tech, w_news, w_ai = 0.2, 0.2, 0.6
|
|
print(f" 🤖 [High Confidence] AI Weight Boosted to 60%")
|
|
else:
|
|
w_tech, w_news, w_ai = 0.4, 0.3, 0.3
|
|
|
|
total_score = (w_tech * tech_score) + (w_news * sentiment_score) + (w_ai * lstm_score)
|
|
|
|
decision = "HOLD"
|
|
|
|
# [신규] 강한 단일 신호 매수 로직 (기준 강화)
|
|
strong_signal = False
|
|
strong_reason = ""
|
|
|
|
if tech_score >= 0.80:
|
|
strong_signal = True
|
|
strong_reason = "Super Strong Technical"
|
|
elif lstm_score >= 0.80 and ai_confidence >= 0.8:
|
|
strong_signal = True
|
|
strong_reason = f"High Confidence AI Buy (Conf: {ai_confidence})"
|
|
elif sentiment_score >= 0.85:
|
|
strong_signal = True
|
|
strong_reason = "Strong News Sentiment"
|
|
|
|
if strong_signal:
|
|
decision = "BUY"
|
|
print(f" 🎯 [{strong_reason}] Overriding to BUY!")
|
|
elif total_score >= 0.60: # (0.5 -> 0.6 상향 조정으로 보수적 접근)
|
|
decision = "BUY"
|
|
elif total_score <= 0.30:
|
|
decision = "SELL"
|
|
|
|
print(f" └─ Scores: Tech={tech_score:.2f} News={sentiment_score:.2f} LSTM={lstm_score:.2f}(Conf:{ai_confidence:.2f}) → Total={total_score:.2f} [{decision}]")
|
|
|
|
# [신규] 변동성(Volatility) 계산
|
|
if len(prices) > 1:
|
|
prices_np = np.array(prices)
|
|
changes = np.diff(prices_np) / prices_np[:-1]
|
|
volatility = np.std(changes) * 100 # 퍼센트 단위
|
|
else:
|
|
volatility = 0.0
|
|
|
|
return {
|
|
"ticker": ticker,
|
|
"score": total_score,
|
|
"tech": tech_score,
|
|
"sentiment": sentiment_score,
|
|
"lstm_score": lstm_score,
|
|
"volatility": volatility,
|
|
"volume_ratio": vol_ratio,
|
|
"prediction": pred_result,
|
|
"decision": decision,
|
|
"current_price": current_price
|
|
}
|