Update app.py

app.py

@@ -1,4 +1,4 @@
-# fusion_analysis_v2.py
+# fusion_analysis_v2_corrected.py
 import pandas as pd
 import firebase_admin
 from firebase_admin import credentials, db
@@ -6,8 +6,6 @@ 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'
@@ -38,17 +36,24 @@ def fetch_data_from_firebase(ref_name):
         
         df = pd.DataFrame.from_dict(data, orient='index')
         
-        # Standardize the timestamp column for merging
+        # Determine the correct timestamp column name
+        timestamp_col = ''
         if 'timestamp_entry' in df.columns: # For agent signals
-            df['timestamp'] = pd.to_datetime(df['timestamp_entry'], utc=True)
+            timestamp_col = 'timestamp_entry'
         elif 'timestamp_published' in df.columns: # For news sentiment
-            df['timestamp'] = pd.to_datetime(df['timestamp_published'], utc=True)
+            timestamp_col = 'timestamp_published'
         else:
             print(f"❌ Timestamp column not found in '{ref_name}'.")
             return pd.DataFrame()
+
+        # --- THIS IS THE FIX ---
+        # Drop any rows where the timestamp column is missing before processing.
+        df.dropna(subset=[timestamp_col], inplace=True)
+        # ---------------------
             
+        df['timestamp'] = pd.to_datetime(df[timestamp_col], utc=True)
         df = df.set_index('timestamp').sort_index()
-        print(f"✅ Fetched {len(df)} records from '{ref_name}'.")
+        print(f"✅ Fetched and validated {len(df)} records from '{ref_name}'.")
         return df
     except Exception as e:
         print(f"❌ Error fetching data from '{ref_name}': {e}")
@@ -70,9 +75,6 @@ def run_fusion_analysis():
     # 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']],
@@ -82,7 +84,6 @@ def run_fusion_analysis():
         tolerance=pd.Timedelta(minutes=30)
     )
     
-    # Drop signals where no recent news was available
     fused_df.dropna(subset=['sentiment'], inplace=True)
     
     if fused_df.empty:
@@ -111,21 +112,20 @@ def run_fusion_analysis():
 
     # 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['pnl'] = pd.to_numeric(evaluated_trades['pnl'])
+        
+        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'
+
         evaluated_trades['alignment'] = evaluated_trades.apply(calculate_alignment, axis=1)
         pnl_by_alignment = evaluated_trades.groupby('alignment')['pnl'].mean()
         print(pnl_by_alignment)