Create app.py

app.py

@@ -0,0 +1,138 @@
+# fusion_analysis_v2.py
+import pandas as pd
+import firebase_admin
+from firebase_admin import credentials, db
+import os
+import json
+
+# --- CONFIGURATION ---
+# IMPORTANT: This script uses the same environment variables as your agents.
+# Ensure FIRESTORE_SA_KEY and FIREBASE_DB_URL are set.
+SA_KEY_JSON = os.environ.get('FIRESTORE_SA_KEY')
+DB_URL = os.environ.get('FIREBASE_DB_URL')
+AGENT_SIGNALS_REF = 'signals_v2'
+NEWS_SENTIMENT_REF = 'news_sentiment'
+
+def initialize_firebase():
+    """Initializes the Firebase app if not already done."""
+    if not firebase_admin._apps:
+        try:
+            cred_dict = json.loads(SA_KEY_JSON)
+            cred = credentials.Certificate(cred_dict)
+            firebase_admin.initialize_app(cred, {'databaseURL': DB_URL})
+            print("✅ Firebase connection successful.")
+            return True
+        except Exception as e:
+            print(f"❌ CRITICAL ERROR - Firebase initialization failed: {e}")
+            return False
+    return True
+
+def fetch_data_from_firebase(ref_name):
+    """Fetches and processes data from a specified Firebase database reference."""
+    try:
+        ref = db.reference(ref_name)
+        data = ref.get()
+        if not data:
+            print(f"⚠️  No data found in '{ref_name}'.")
+            return pd.DataFrame()
+        
+        df = pd.DataFrame.from_dict(data, orient='index')
+        
+        # Standardize the timestamp column for merging
+        if 'timestamp_entry' in df.columns: # For agent signals
+            df['timestamp'] = pd.to_datetime(df['timestamp_entry'], utc=True)
+        elif 'timestamp_published' in df.columns: # For news sentiment
+            df['timestamp'] = pd.to_datetime(df['timestamp_published'], utc=True)
+        else:
+            print(f"❌ Timestamp column not found in '{ref_name}'.")
+            return pd.DataFrame()
+            
+        df = df.set_index('timestamp').sort_index()
+        print(f"✅ Fetched {len(df)} records from '{ref_name}'.")
+        return df
+    except Exception as e:
+        print(f"❌ Error fetching data from '{ref_name}': {e}")
+        return pd.DataFrame()
+
+def run_fusion_analysis():
+    """Main function to fetch, fuse, and analyze the data."""
+    if not initialize_firebase():
+        return
+
+    # 1. Fetch both datasets
+    agent_df = fetch_data_from_firebase(AGENT_SIGNALS_REF)
+    news_df = fetch_data_from_firebase(NEWS_SENTIMENT_REF)
+
+    if agent_df.empty or news_df.empty:
+        print("\nCould not proceed with fusion. One or both data sources are empty.")
+        return
+
+    # 2. Fuse the datasets using time
+    print("\n--- Fusing Agent Decisions with News Sentiment ---")
+    
+    # This is the core logic: For each agent signal, find the most recent news
+    # that was published right before it (within a 30-minute tolerance).
+    # This checks if "news was available" and relevant at the time of the decision.
+    fused_df = pd.merge_asof(
+        left=agent_df,
+        right=news_df[['sentiment', 'confidence_score', 'headline']],
+        left_index=True,
+        right_index=True,
+        direction='backward',
+        tolerance=pd.Timedelta(minutes=30)
+    )
+    
+    # Drop signals where no recent news was available
+    fused_df.dropna(subset=['sentiment'], inplace=True)
+    
+    if fused_df.empty:
+        print("\nNo agent signals were found to have occurred within 30 minutes of a news headline. Analysis cannot proceed.")
+        return
+
+    print(f"✅ Fusion complete. Matched {len(fused_df)} agent signals with recent news headlines.")
+
+    # --- 🧠 THE INTELLIGENCE LABORATORY 🧠 ---
+    print("\n--- Advanced Analysis Results ---")
+
+    # Analysis 1: How does news sentiment correlate with the agent's actions?
+    print("\n[Analysis 1: Agent Action vs. News Sentiment]")
+    action_sentiment_crosstab = pd.crosstab(fused_df['action'], fused_df['sentiment'])
+    print(action_sentiment_crosstab)
+    print("Insight: This shows if the agent tends to BUY during 'Bullish' news and SELL during 'Bearish' news.")
+
+    # Analysis 2: Let's bring in the agent's own understanding of the market regime.
+    print("\n[Analysis 2: Action vs. Sentiment, Segmented by Market Regime]")
+    regime_crosstab = pd.crosstab(
+        [fused_df['market_regime'], fused_df['action']],
+        fused_df['sentiment']
+    )
+    print(regime_crosstab)
+    print("Insight: This is powerful. It might reveal if the agent only trusts 'Bullish' news when it already believes the market is 'TRENDING'.")
+
+    # Analysis 3: Does trading in alignment with high-confidence news lead to better outcomes?
+    print("\n[Analysis 3: Trade PnL vs. News Sentiment Alignment]")
+    # Filter for signals that have already been evaluated (have a PnL value)
+    evaluated_trades = fused_df.dropna(subset=['pnl'])
+    evaluated_trades['pnl'] = pd.to_numeric(evaluated_trades['pnl'])
+    
+    # Create an 'alignment' score
+    def calculate_alignment(row):
+        if (row['action'] == 'BUY' and row['sentiment'] == 'Bullish') or \
+           (row['action'] == 'SELL' and row['sentiment'] == 'Bearish'):
+            return 'Aligned'
+        elif (row['action'] == 'BUY' and row['sentiment'] == 'Bearish') or \
+             (row['action'] == 'SELL' and row['sentiment'] == 'Bullish'):
+            return 'Misaligned'
+        return 'Neutral'
+
+    if not evaluated_trades.empty:
+        evaluated_trades['alignment'] = evaluated_trades.apply(calculate_alignment, axis=1)
+        pnl_by_alignment = evaluated_trades.groupby('alignment')['pnl'].mean()
+        print(pnl_by_alignment)
+        print("Insight: This is the ultimate test. It shows if trades that 'agree' with the news are, on average, more profitable.")
+    else:
+        print("No evaluated trades with PnL found to analyze performance.")
+
+
+if __name__ == "__main__":
+    run_fusion_analysis()