Scripts calcul hyperopt multiple

tools/detect_regime.py

@@ -33,44 +33,67 @@ if 'close' not in df.columns or 'timestamp' not in df.columns:
     sys.exit(1)
 
 # --- paramètres ---
-SMOOTH_WIN = 5   # EMA pour lisser la pente
-NUM_TOP = 3      # nombre de segments à garder par classe
+SMOOTH_WIN = 10   # EMA pour lisser la pente
+NUM_TOP = 1      # nombre de segments à garder par classe
 
 # --- charger les données ---
 df['timestamp'] = pd.to_datetime(df['timestamp'], errors='coerce')
 
+# filtrer uniquement 2024 et 2025
+df = df[df['timestamp'].dt.year.isin([2024, 2025])]
+
 # --- calcul SMA14 ---
-df['sma14'] = ta.trend.sma_indicator(df['close'], 14)
+df['sma'] = talib.SMA(df, timeperiod=20) #ta.trend.sma_indicator(df['close'], 14)
 
 # --- pente brute ---
-df['slope'] = df['sma14'].diff()
+df['slope'] = df['sma'].diff()
 
 # --- lissage EMA ---
 df['slope_smooth'] = df['slope'].ewm(span=SMOOTH_WIN, adjust=False).mean()
 
+#df["slope_smooth"] = savgol_filter(df["slope_smooth"], window_length=21, polyorder=3)
+
 # --- normalisation relative ---
 df['slope_norm'] = df['slope_smooth'] / df['close']
-df['slope_norm'].fillna(0, inplace=True)
+# df['slope_norm'].fillna(0, inplace=True)
+df['slope_norm'] = df['slope_norm'].fillna(0)
+
 
 # --- classification dynamique via quantiles ---
-q = df['slope_norm'].quantile([0.07, 0.21, 0.35, 0.65, 0.79, 0.93]).values
-q1, q2, q3, q4, q5, q6 = q
+
+q = df['slope_norm'].quantile([0.125, 0.375, 0.625, 0.875]).values
+q1, q2, q3, q4 = q
 
 def classify(v):
     if v <= q1:
-        return 'BR3'
-    elif v <= q2:
         return 'BR2'
-    elif v <= q3:
+    elif v <= q2:
         return 'BR1'
-    elif v <= q4:
+    elif v <= q3:
         return 'RG'
-    elif v <= q5:
+    elif v <= q4:
         return 'BU1'
-    elif v <= q6:
-        return 'BU2'
     else:
-        return 'BU3'
+        return 'BU2'
+
+# q = df['slope_norm'].quantile([0.7, 0.21, 0.35, 0.65, 0.79, 0.93]).values
+# q1, q2, q3, q4, q5, q6 = q
+#
+# def classify(v):
+#     if v <= q1:
+#         return 'BR3'
+#     elif v <= q2:
+#         return 'BR2'
+#     elif v <= q3:
+#         return 'BR1'
+#     elif v <= q4:
+#         return 'RG'
+#     elif v <= q5:
+#         return 'BU1'
+#     elif v <= q6:
+#         return 'BU2'
+#     else:
+#         return 'BU3'
 
 df['trend_class'] = df['slope_norm'].apply(classify)
 
@@ -100,7 +123,7 @@ if trend is not None and start_idx is not None:
 
 # --- extraire les 5 plus longs segments par classe ---
 top_segments_by_class = {}
-for cls in ['BR3','BR2','BR1','RG','BU1','BU2','BU3']:
+for cls in ['BR2','BR1','RG','BU1','BU2']:
     cls_segments = [(t,s,e) for t,s,e in segments if t==cls]
     # calcul de la durée
     cls_segments = [(t,s,e,(e-s).total_seconds()) for t,s,e in cls_segments]