use sample weights in keras model

pandas_ml_utils/model/models.py

@@ -287,16 +287,12 @@ def fit(self,
             sample_weight_train: np.ndarray, sample_weight_test: np.ndarray) -> float:
         fitter_args = suitable_kwargs(self.keras_model.fit, **self.kwargs)
 
-        #if self.sample_weight_column:
-            # TODO now we can access the raw data frame from features and labels extractor
-            #  so if we have a column defined for class weight we could provide the weighs for each sample to the fitter!
-            # FIXME doen not work yet: fitter_args["sample_weight"] = self.features_and_labels[self.sample_weight_column].values
-
         if "verbose" in self.kwargs and self.kwargs["verbose"] > 0:
             print(f'pass args to fit: {fitter_args}')
 
         fit_history = self._exec_within_session(self.keras_model.fit,
                                                 x, y,
+                                                sample_weight=sample_weight_train.reshape((len(x), )),
                                                 epochs=self.epochs,
                                                 validation_data=(x_val, y_val),
                                                 callbacks=[cb() for cb in self.callbacks],
@@ -437,8 +433,10 @@ def fit(self,
             index = range(pos, pos + len(labels))
             target_y = y[:,index]
             target_y_val = y_val[:,index]
+            target_w = sample_weight_train[:,index]
+            target_w_val = sample_weight_test[:,index]
             _log.info(f"fit model for target {target}")
-            losses.append(self.models[target].fit(x, target_y, x_val, target_y_val, sample_weight_train, sample_weight_test))
+            losses.append(self.models[target].fit(x, target_y, x_val, target_y_val, target_w, target_w_val))
             pos += len(labels)
 
         losses = np.array(losses)

test/unit_tests/model/test__model.py

@@ -81,8 +81,9 @@ def keras_model_provider(optimizer='adam'):
         np.testing.assert_array_almost_equal(model1.get_weights(), model2.get_weights())
 
         """and after we fit one model"""
-        loss = model2.fit(np.array([0.1, 0.01]), np.array([0.1, 0.01]), np.array([0.1, 0.01]), np.array([0.1, 0.01]),
-                          [0,1], [2,3])
+        loss = model2.fit(np.array([0.1, 0.01]), np.array([0.1, 0.01]),
+                          np.array([0.1, 0.01]), np.array([0.1, 0.01]),
+                          np.array([1,1]), np.array([1,1]))
 
         """then"""
         self.assertIsNotNone(loss)