TensorFlow

Zeus_TensorFlow.py

@@ -108,17 +108,17 @@ def normalize(df):
     return df
 
 
-class Zeus_LGBMRegressor(IStrategy):
+class Zeus_TensorFlow(IStrategy):
     startup_candle_count = 24 * 12
 
     # Machine Learning
-    model_indicators = []
-    indicator_target = 'percent12'
-
     model = None
+    model_indicators = []
+    indicator_target = 'mid_smooth_5'
     # Tensorflow
     lookback = 60
     future_steps = 12
+    y_no_scale = False
 
     path = f"user_data/plots/"
 
@@ -3040,8 +3040,8 @@ class Zeus_LGBMRegressor(IStrategy):
 
         # 7) Sauvegarde
         self.model.save(f"{self.path}/lstm_model.keras")
-        np.save(f"{self.path}/lstm_scaler_X.npy", self.scaler_X.data_max_)
-        np.save(f"{self.path}/lstm_scaler_y.npy", self.scaler_y.data_max_)
+        # np.save(f"{self.path}/lstm_scaler_X.npy", self.scaler_X.data_max_)
+        # np.save(f"{self.path}/lstm_scaler_y.npy", self.scaler_y.data_max_)
 
     def tensorFlowPrepareDataFrame(self, dataframe, future_steps, lookback):
         target = self.indicator_target
@@ -3062,8 +3062,13 @@ class Zeus_LGBMRegressor(IStrategy):
         # 4) Normalisation
         self.scaler_X = MinMaxScaler()
         X_scaled = self.scaler_X.fit_transform(X_values)
-        self.scaler_y = MinMaxScaler()
-        y_scaled = self.scaler_y.fit_transform(y_values)
+
+        if self.y_no_scale:
+            y_scaled = y_values
+        else:
+            self.scaler_y = MinMaxScaler()
+            y_scaled = self.scaler_y.fit_transform(y_values)
+
         # 5) Création des fenêtres glissantes
         X_seq = []
         y_seq = []
@@ -3152,10 +3157,12 @@ class Zeus_LGBMRegressor(IStrategy):
             X_seq.append(X_scaled[i:i + lookback])
         X_seq = np.array(X_seq)
 
-
         # prédictions
         y_pred_scaled = model.predict(X_seq, verbose=0).flatten()
-        y_pred = self.scaler_y.inverse_transform(y_pred_scaled.reshape(-1, 1)).flatten()
+        if self.y_no_scale:
+            y_pred = y_pred_scaled
+        else:
+            y_pred = self.scaler_y.inverse_transform(y_pred_scaled.reshape(-1, 1)).flatten()
         # alignement avec les données
         preds = [np.nan] * len(dataframe)
         start = lookback