Upload 2 files

app.py

@@ -240,14 +240,17 @@ def attach_market_state(payload: dict) -> dict:
     try:
         models_dir = Path(__file__).resolve().parent / "models"
         mfe_out = models_dir / "nifty_opening_mfe_regressor" / "outputs"
+        data_dir = Path(__file__).resolve().parent / "data"
         if (mfe_out / "summary.json").exists():
             mfe_summary_fallback = json.loads((mfe_out / "summary.json").read_text(encoding="utf-8"))
         if (mfe_out / "latest_prediction.csv").exists():
             row = pd.read_csv(mfe_out / "latest_prediction.csv").iloc[-1].to_dict()
             mfe_latest_fallback = {k: (None if pd.isna(v) else v) for k, v in row.items()}
+            
+        hist_records = []
+        import numpy as np
         if (mfe_out / "test_predictions.csv").exists():
             hist_df = pd.read_csv(mfe_out / "test_predictions.csv")
-            hist_records = []
             for _, r in hist_df.iterrows():
                 try:
                     dt = str(r["date"])
@@ -258,6 +261,9 @@ def attach_market_state(payload: dict) -> dict:
                     act_lo = float(r["day_low"])
                     hist_records.append({
                         "date": dt,
+                        "first5_close": f5c,
+                        "predicted_up_points": pred_up,
+                        "predicted_down_points": pred_dn,
                         "actual_high": act_hi,
                         "predicted_high": f5c + pred_up,
                         "actual_low": act_lo,
@@ -265,7 +271,74 @@ def attach_market_state(payload: dict) -> dict:
                     })
                 except Exception:
                     continue
-            mfe_history_fallback = hist_records
+        
+        # Load live history if it exists
+        if (mfe_out / "mfe_live_history.csv").exists():
+            live_df = pd.read_csv(mfe_out / "mfe_live_history.csv")
+            daily_df = None
+            if (data_dir / "nifty50_1d.parquet").exists():
+                daily_df = pd.read_parquet(data_dir / "nifty50_1d.parquet")
+                daily_df["date"] = pd.to_datetime(daily_df["date"]).dt.strftime("%Y-%m-%d")
+                daily_df = daily_df.set_index("date")
+            
+            for _, r in live_df.iterrows():
+                try:
+                    dt = str(r["input_date"])
+                    if daily_df is not None and dt in daily_df.index:
+                        # Extract as scalar float using .iloc[0] or .item() in case of duplicates
+                        act_hi_raw = daily_df.loc[dt, "high"]
+                        act_lo_raw = daily_df.loc[dt, "low"]
+                        act_hi = float(act_hi_raw.iloc[0] if isinstance(act_hi_raw, pd.Series) else act_hi_raw)
+                        act_lo = float(act_lo_raw.iloc[0] if isinstance(act_lo_raw, pd.Series) else act_lo_raw)
+                        f5c = float(r["first5_close"])
+                        pred_up = float(r["predicted_up_points"])
+                        pred_dn = float(r["predicted_down_points"])
+                        hist_records.append({
+                            "date": dt,
+                            "first5_close": f5c,
+                            "predicted_up_points": pred_up,
+                            "predicted_down_points": pred_dn,
+                            "actual_high": act_hi,
+                            "predicted_high": f5c + pred_up,
+                            "actual_low": act_lo,
+                            "predicted_low": f5c - pred_dn
+                        })
+                except Exception as ex:
+                    print(f"Error appending live row: {ex}")
+                    continue
+                    
+        mfe_history_fallback = hist_records
+        
+        # Recalculate RMSE and MAE over the combined history
+        if hist_records:
+            up_errors = []
+            down_errors = []
+            for r in hist_records:
+                pred_up_pts = r["predicted_up_points"]
+                pred_dn_pts = r["predicted_down_points"]
+                act_up_pts = r["actual_high"] - r["first5_close"]
+                act_dn_pts = r["first5_close"] - r["actual_low"]
+                up_errors.append(act_up_pts - pred_up_pts)
+                down_errors.append(act_dn_pts - pred_dn_pts)
+                
+            up_errors = np.array(up_errors)
+            down_errors = np.array(down_errors)
+            
+            up_rmse = float(np.sqrt(np.mean(up_errors**2)))
+            up_mae = float(np.mean(np.abs(up_errors)))
+            down_rmse = float(np.sqrt(np.mean(down_errors**2)))
+            down_mae = float(np.mean(np.abs(down_errors)))
+            
+            if "up" not in mfe_summary_fallback:
+                mfe_summary_fallback["up"] = {}
+            if "down" not in mfe_summary_fallback:
+                mfe_summary_fallback["down"] = {}
+                
+            mfe_summary_fallback["up"]["test_rmse_points"] = up_rmse
+            mfe_summary_fallback["up"]["test_mae_points"] = up_mae
+            mfe_summary_fallback["down"]["test_rmse_points"] = down_rmse
+            mfe_summary_fallback["down"]["test_mae_points"] = down_mae
+            
     except Exception as exc:
         print(f"Fallback MFE load failed: {exc}", flush=True)
 

runtime.py

@@ -61,6 +61,7 @@ MFE_SUMMARY_PATH = MFE_OUTPUT_DIR / "summary.json"
 MFE_LATEST_PATH = MFE_OUTPUT_DIR / "latest_prediction.csv"
 MFE_TEST_PREDICTIONS_PATH = MFE_OUTPUT_DIR / "test_predictions.csv"
 MFE_MODEL_PATH = MFE_OUTPUT_DIR / "nifty_opening_mfe_regressor.joblib"
+MFE_LIVE_HISTORY_PATH = MFE_OUTPUT_DIR / "mfe_live_history.csv"
 TPLUS1_MODEL_PATH = MODEL_DIR / "nifty_1420_tplus1_logistic_model.joblib"
 TPLUS1_LATEST_PATH = MODEL_DIR / "tplus1_latest_prediction.csv"
 TPLUS1_SUMMARY_PATH = MODEL_DIR / "tplus1_summary.json"
@@ -1002,6 +1003,20 @@ def refresh_mfe_prediction(session_date: date | None = None) -> dict[str, Any]:
         "predicted_down_points": pred_down,
     }
     pd.DataFrame([out]).to_csv(MFE_LATEST_PATH, index=False)
+    
+    # Append to live history
+    live_df = pd.DataFrame([out])
+    if MFE_LIVE_HISTORY_PATH.exists():
+        try:
+            existing = pd.read_csv(MFE_LIVE_HISTORY_PATH)
+            # Avoid duplicates if refreshed multiple times in the same session
+            existing = existing[existing["input_date"] != out["input_date"]]
+            pd.concat([existing, live_df], ignore_index=True).to_csv(MFE_LIVE_HISTORY_PATH, index=False)
+        except Exception:
+            live_df.to_csv(MFE_LIVE_HISTORY_PATH, index=False)
+    else:
+        live_df.to_csv(MFE_LIVE_HISTORY_PATH, index=False)
+        
     clear_dashboard_payload_cache()
     return out