dquant 1.2.2__tar.gz → 1.2.4__tar.gz

{dquant-1.2.2/src/DQuant.egg-info → dquant-1.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dquant
-Version: 1.2.2
+Version: 1.2.4
 Summary: DQuant is an open-source Python library for automated volatility forecasting of financial time series. It handles all stages of model construction, from raw prices to the final forecast.
 Author: Denis Makarov
 Project-URL: Homepage, https://dquant.space

{dquant-1.2.2 → dquant-1.2.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "dquant"
-version = "1.2.2"
+version = "1.2.4"
 authors = [
   { name="Denis Makarov" },
 ]

{dquant-1.2.2 → dquant-1.2.4/src/DQuant.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dquant
-Version: 1.2.2
+Version: 1.2.4
 Summary: DQuant is an open-source Python library for automated volatility forecasting of financial time series. It handles all stages of model construction, from raw prices to the final forecast.
 Author: Denis Makarov
 Project-URL: Homepage, https://dquant.space

{dquant-1.2.2 → dquant-1.2.4}/src/dquant/metrics.py

@@ -2,7 +2,7 @@ import numpy as np
 
 
 def qlike_score(y_true, y_pred):
-    sigma2_true = y_true ** 2
+    sigma2_true = y_true
     sigma2_pred = np.maximum(y_pred, 1e-10)
     return np.mean(np.log(sigma2_pred) + sigma2_true / sigma2_pred)
 

{dquant-1.2.2 → dquant-1.2.4}/src/dquant/models.py

@@ -557,8 +557,14 @@ class FichEn:
         X_train, X_test, y_train, y_test = train_test_split(x, y, test_size=0.2, shuffle=False, random_state=42)
         X_scaled = self.scaler.fit_transform(X_train)
         X_test_scaled = self.scaler.transform(X_test)
-        Y_scaled = self.scaler_y.fit_transform(y_train)
-        Y_test_scaled = self.scaler_y.transform(y_test)
+        #Y_scaled = self.scaler_y.fit_transform(y_train)
+        #Y_test_scaled = self.scaler_y.transform(y_test)
+        """if hasattr(y, 'ndim') and y.ndim == 1:
+            self.X_shape = 1
+            #y_2d = y.values.reshape(-1, 1) if hasattr(y, 'values') else y.reshape(-1, 1)
+        else:
+            self.X_shape = y.shape[1]"""
+
         self.X_shape = x.shape[1]
 
         self.train_errors = []
@@ -594,17 +600,18 @@ class FichEn:
                     horizon_list = list(range(horizon))
                 else:
                     horizon_list = horizon
-                if len(Y_scaled.shape) == 2 and Y_scaled.shape[1] > 0:
+                if len(y_train.shape) == 2 and y_train.shape[1] > 0:
                     for h_idx, h in enumerate(horizon_list):
-                        if h_idx >= Y_scaled.shape[1]:
+                        if h_idx >= y_train.shape[1]:
                             self.dquantprint(f"Warning: horizon {h} extends beyond y, skipping")
                             continue
 
-                        y_h = Y_scaled.iloc[:, h_idx] if hasattr(Y_scaled, 'iloc') else Y_scaled[:, h_idx]
+                        y_h = y_train.iloc[:, h_idx] if hasattr(y_train, 'iloc') else y_train[:, h_idx]
 
                         valid_mask = ~pd.isna(y_h) if hasattr(y_h, 'isna') else ~np.isnan(y_h)
                         X_h = X_scaled[valid_mask]
                         y_h_clean = y_h[valid_mask]
+                        #y_h_clean_orig = self.scaler_y.inverse_transform(y_h_clean.reshape(-1, 1)).ravel()
 
                         if i != 1:
                             self.models[h_idx].set_params(n_estimators=i)
@@ -615,29 +622,38 @@ class FichEn:
                             model.fit(X_h, y_h_clean)
                             self.models.append(model)
 
-                        y_h_v = Y_test_scaled.iloc[:, h_idx] if hasattr(Y_test_scaled, 'iloc') else Y_test_scaled[:, h_idx]
+                        y_h_v = y_test.iloc[:, h_idx] if hasattr(y_test, 'iloc') else y_test[:, h_idx]
 
                         valid_mask = ~pd.isna(y_h_v) if hasattr(y_h_v, 'isna') else ~np.isnan(y_h_v)
                         X_h_v = X_test_scaled[valid_mask]
                         y_h_v_clean = y_h_v[valid_mask]
+                        #y_h_v_clean_orig = self.scaler_y.inverse_transform(y_h_v_clean.reshape(-1, 1)).ravel()
                         if i != 1:
-                            t_error += mean_squared_error(y_h_clean, self.models[h_idx].predict(X_h))
-                            v_error += mean_squared_error(y_h_v_clean, self.models[h_idx].predict(X_h_v))
-                            t_mae += mean_absolute_error(y_h_clean, self.models[h_idx].predict(X_h))
-                            v_mae += mean_absolute_error(y_h_v_clean, self.models[h_idx].predict(X_h_v))
-                            t_qlike += qlike_score(y_h_clean, self.models[h_idx].predict(X_h))
-                            v_qlike += qlike_score(y_h_v_clean, self.models[h_idx].predict(X_h_v))
-                            t_r2 += r2_score(y_h_clean, self.models[h_idx].predict(X_h))
-                            v_r2 += r2_score(y_h_v_clean, self.models[h_idx].predict(X_h_v))
+                            pred_train = self.models[h_idx].predict(X_h)
+                            pred_val = self.models[h_idx].predict(X_h_v)
+                            #pred_train_orig = self.scaler_y.inverse_transform(pred_train.reshape(-1, 1)).ravel()
+                            #pred_val_orig = self.scaler_y.inverse_transform(pred_val.reshape(-1, 1)).ravel()
+                            t_error += mean_squared_error(y_h_clean, pred_train)
+                            v_error += mean_squared_error(y_h_v_clean, pred_val)
+                            t_mae += mean_absolute_error(y_h_clean, pred_train)
+                            v_mae += mean_absolute_error(y_h_v_clean, pred_val)
+                            t_qlike += qlike_score(y_h_clean, pred_train)
+                            v_qlike += qlike_score(y_h_v_clean, pred_val)
+                            t_r2 += r2_score(y_h_clean, pred_train)
+                            v_r2 += r2_score(y_h_v_clean, pred_val)
                         else:
-                            t_error += mean_squared_error(y_h_clean, model.predict(X_h))
-                            v_error += mean_squared_error(y_h_v_clean, model.predict(X_h_v))
-                            t_mae += mean_absolute_error(y_h_clean, model.predict(X_h))
-                            v_mae += mean_absolute_error(y_h_v_clean, model.predict(X_h_v))
-                            t_qlike += qlike_score(y_h_clean, model.predict(X_h))
-                            v_qlike += qlike_score(y_h_v_clean, model.predict(X_h_v))
-                            t_r2 += r2_score(y_h_clean, model.predict(X_h))
-                            v_r2 += r2_score(y_h_v_clean, model.predict(X_h_v))
+                            pred_train = model.predict(X_h)
+                            pred_val = model.predict(X_h_v)
+                            #pred_train_orig = self.scaler_y.inverse_transform(pred_train.reshape(-1, 1)).ravel()
+                            #pred_val_orig = self.scaler_y.inverse_transform(pred_val.reshape(-1, 1)).ravel()
+                            t_error += mean_squared_error(y_h_clean, pred_train)
+                            v_error += mean_squared_error(y_h_v_clean, pred_val)
+                            t_mae += mean_absolute_error(y_h_clean, pred_train)
+                            v_mae += mean_absolute_error(y_h_v_clean, pred_val)
+                            t_qlike += qlike_score(y_h_clean, pred_train)
+                            v_qlike += qlike_score(y_h_v_clean, pred_val)
+                            t_r2 += r2_score(y_h_clean, pred_train)
+                            v_r2 += r2_score(y_h_v_clean, pred_val)
 
 
                 var_test_error = float(t_error)/horizon
@@ -738,7 +754,7 @@ class FichEn:
                     pred_array = pred_array.T
                 elif pred_array.shape[0] > 1 and pred_array.shape[1] == 30:
                     pred_array = pred_array[0:1, :]
-            predictions = self.scaler_y.inverse_transform(pred_array).flatten()
+            predictions = pred_array.flatten()
 
             if show:
                 epsilon = 1e-10
@@ -782,7 +798,7 @@ class FichEn:
                     pred_array = pred_array.T
                 elif pred_array.shape[0] > 1 and pred_array.shape[1] == 30:
                     pred_array = pred_array[0:1, :]
-            predictions = self.scaler_y.inverse_transform(pred_array).flatten()
+            predictions = pred_array.flatten()
 
             if show:
                 epsilon = 1e-10
@@ -816,13 +832,13 @@ class FichEn:
         mean_str = ','.join(str(x) for x in scaler_data['mean'])
         std_str = ','.join(str(x) for x in scaler_data['std'])
 
-        scaler_data_y = {
+        """scaler_data_y = {
             "mean": self.scaler_y.mean_.tolist() if self.scaler_y.mean_ is not None else [],
             "std": self.scaler_y.scale_.tolist() if self.scaler_y.scale_ is not None else [],
             "var": self.scaler_y.var_.tolist() if self.scaler_y.var_ is not None else []
         }
         mean_str_y = ','.join(str(x) for x in scaler_data_y['mean'])
-        std_str_y = ','.join(str(x) for x in scaler_data_y['std'])
+        std_str_y = ','.join(str(x) for x in scaler_data_y['std'])"""
 
 
         os.makedirs(name, exist_ok=True)
@@ -854,8 +870,8 @@ class FichEn:
             f.write(f"double mean_[] = {{{mean_str}}};\n\n")
             f.write(f"double std_[] = {{{std_str}}};\n\n")
 
-            f.write(f"double mean_y[] = {{{mean_str_y}}};\n\n")
-            f.write(f"double std_y[] = {{{std_str_y}}};\n\n")
+            #f.write(f"double mean_y[] = {{{mean_str_y}}};\n\n")
+            #f.write(f"double std_y[] = {{{std_str_y}}};\n\n")
 
             f.write("//--- indicator buffers\n")
             f.write("double past_vol[];\n")
@@ -1469,7 +1485,7 @@ class VolClustGB(FichEn):
         self.output = output
         self.models = []
         self.scaler = StandardScaler()
-        self.scaler_y = StandardScaler()
+        #self.scaler_y = StandardScaler()
         self.X_shape = 0
         self.is_fitted = False
         self.onnx_load = False
@@ -1520,9 +1536,9 @@ class VolClustGB(FichEn):
                 scaler_path = os.path.join(name, f"{name}_scaler.pkl")
                 joblib.dump(self.scaler, scaler_path)
 
-            if hasattr(self, 'scaler_y'):
+            """if hasattr(self, 'scaler_y'):
                 scaler_path = os.path.join(name, f"{name}_scaler_y.pkl")
-                joblib.dump(self.scaler_y, scaler_path)
+                joblib.dump(self.scaler_y, scaler_path)"""
 
             for i in range(len(self.models)):
                 onx = convert_sklearn(self.models[i], initial_types=initial_type, target_opset=12)
@@ -1544,10 +1560,10 @@ class VolClustGB(FichEn):
                 joblib.dump(self.scaler, scaler_path)
                 self.dquantprint(f"Scaler is saved in {scaler_path}")
 
-            if hasattr(self, 'scaler_y') and self.scaler_y is not None:
+            """if hasattr(self, 'scaler_y') and self.scaler_y is not None:
                 scaler_path = os.path.join(onnx_dir, f"{name}_scaler_y.pkl")
                 joblib.dump(self.scaler_y, scaler_path)
-                self.dquantprint(f"Scalery is saved in {scaler_path}")
+                self.dquantprint(f"Scalery is saved in {scaler_path}")"""
 
             for i in range(len(self.models)):
                 onx = convert_sklearn(self.models[i], initial_types=initial_type, target_opset=12)
@@ -1589,10 +1605,10 @@ class VolClustGB(FichEn):
             scaler_path = os.path.join(name, scaler_files[0])
             self.scaler = joblib.load(scaler_path)
 
-        scaler_files = [f for f in os.listdir(name) if f.endswith('_scaler_y.pkl')]
+        """scaler_files = [f for f in os.listdir(name) if f.endswith('_scaler_y.pkl')]
         if scaler_files:
-            scaler_path = os.path.join(name, scaler_files[0])
-            self.scaler_y = joblib.load(scaler_path)
+            scaler_path = os.path.join(name, scaler_files[0])"""
+            #self.scaler_y = joblib.load(scaler_path)
 
         model_files = [f for f in os.listdir(name) if f.endswith('.onnx')]
 
@@ -1631,7 +1647,7 @@ class VolClustXGB(FichEn):
         self.output = output
         self.models = []
         self.scaler = StandardScaler()
-        self.scaler_y = StandardScaler()
+        #self.scaler_y = StandardScaler()
         self.X_shape = 0
         self.is_fitted = False
         self.onnx_load = False
@@ -1701,9 +1717,9 @@ class VolClustXGB(FichEn):
                 scaler_path = os.path.join(name, f"{name}_scaler.pkl")
                 joblib.dump(self.scaler, scaler_path)
 
-            if hasattr(self, 'scaler_y'):
+            """if hasattr(self, 'scaler_y'):
                 scaler_path = os.path.join(name, f"{name}_scaler_y.pkl")
-                joblib.dump(self.scaler_y, scaler_path)
+                joblib.dump(self.scaler_y, scaler_path)"""
 
             for i in range(len(self.models)):
                 onx = onnxmltools.convert_xgboost(self.models[i], initial_types=initial_type, target_opset=12)
@@ -1723,10 +1739,10 @@ class VolClustXGB(FichEn):
                 joblib.dump(self.scaler, scaler_path)
                 self.dquantprint(f"Scaler is saved in {scaler_path}")
 
-            if hasattr(self, 'scaler_y') and self.scaler_y is not None:
+            """if hasattr(self, 'scaler_y') and self.scaler_y is not None:
                 scaler_path = os.path.join(onnx_dir, f"{name}_scaler_y.pkl")
                 joblib.dump(self.scaler_y, scaler_path)
-                self.dquantprint(f"Scalery is saved in {scaler_path}")
+                self.dquantprint(f"Scalery is saved in {scaler_path}")"""
 
             for i in range(len(self.models)):
                 onx = onnxmltools.convert_xgboost(self.models[i], initial_types=initial_type, target_opset=9)
@@ -1768,10 +1784,10 @@ class VolClustXGB(FichEn):
             scaler_path = os.path.join(name, scaler_files[0])
             self.scaler = joblib.load(scaler_path)
 
-        scaler_files = [f for f in os.listdir(name) if f.endswith('_scaler_y.pkl')]
+        """scaler_files = [f for f in os.listdir(name) if f.endswith('_scaler_y.pkl')]
         if scaler_files:
             scaler_path = os.path.join(name, scaler_files[0])
-            self.scaler_y = joblib.load(scaler_path)
+            #self.scaler_y = joblib.load(scaler_path)"""
 
         model_files = [f for f in os.listdir(name) if f.endswith('.onnx')]
 
@@ -1808,7 +1824,7 @@ class VolClustLightGBM(FichEn):
         self.output = output
         self.models = []
         self.scaler = StandardScaler()
-        self.scaler_y = StandardScaler()
+        #self.scaler_y = StandardScaler()
         self.X_shape = 0
         self.is_fitted = False
         self.onnx_load = False
@@ -1880,9 +1896,9 @@ class VolClustLightGBM(FichEn):
                 scaler_path = os.path.join(name, f"{name}_scaler.pkl")
                 joblib.dump(self.scaler, scaler_path)
 
-            if hasattr(self, 'scaler_y'):
+            """if hasattr(self, 'scaler_y'):
                 scaler_path = os.path.join(name, f"{name}_scaler_y.pkl")
-                joblib.dump(self.scaler_y, scaler_path)
+                joblib.dump(self.scaler_y, scaler_path)"""
 
             for i in range(len(self.models)):
                 onx = onnxmltools.convert_lightgbm(self.models[i], initial_types=initial_type, zipmap=False,
@@ -1902,10 +1918,10 @@ class VolClustLightGBM(FichEn):
                 joblib.dump(self.scaler, scaler_path)
                 self.dquantprint(f"Scaler is saved in {scaler_path}")
 
-            if hasattr(self, 'scaler_y') and self.scaler_y is not None:
+            """if hasattr(self, 'scaler_y') and self.scaler_y is not None:
                 scaler_path = os.path.join(onnx_dir, f"{name}_scaler_y.pkl")
                 joblib.dump(self.scaler_y, scaler_path)
-                self.dquantprint(f"Scalery is saved in {scaler_path}")
+                self.dquantprint(f"Scalery is saved in {scaler_path}")"""
 
             for i in range(len(self.models)):
                 onx = onnxmltools.convert_lightgbm(self.models[i], initial_types=initial_type, zipmap=False,
@@ -1947,10 +1963,10 @@ class VolClustLightGBM(FichEn):
             scaler_path = os.path.join(name, scaler_files[0])
             self.scaler = joblib.load(scaler_path)
 
-        scaler_files = [f for f in os.listdir(name) if f.endswith('_scaler_y.pkl')]
+        """scaler_files = [f for f in os.listdir(name) if f.endswith('_scaler_y.pkl')]
         if scaler_files:
             scaler_path = os.path.join(name, scaler_files[0])
-            self.scaler_y = joblib.load(scaler_path)
+            #self.scaler_y = joblib.load(scaler_path)"""
 
         model_files = [f for f in os.listdir(name) if f.endswith('.onnx')]