Fixes

rdt/transformers/numerical.py

@@ -463,7 +463,6 @@ def _fit(self, data):
             n_components=self.max_clusters,
             weight_concentration_prior_type='dirichlet_process',
             weight_concentration_prior=0.001,
-            n_init=1,
             random_state=self._get_current_random_seed()
         )
 
@@ -494,10 +493,12 @@ def _transform(self, data):
 
         data = data.reshape((len(data), 1))
         means = self._bgm_transformer.means_.reshape((1, self.max_clusters))
-
+        means = means[:, self.valid_component_indicator]
         stds = np.sqrt(self._bgm_transformer.covariances_).reshape((1, self.max_clusters))
+        stds = stds[:, self.valid_component_indicator]
+
+        # Multiply stds by 4 so that a value will be in the range [-1,1] with 99.99% probability
         normalized_values = (data - means) / (self.STD_MULTIPLIER * stds)
-        normalized_values = normalized_values[:, self.valid_component_indicator]
         component_probs = self._bgm_transformer.predict_proba(data)
         component_probs = component_probs[:, self.valid_component_indicator]
 
@@ -524,7 +525,8 @@ def _reverse_transform_helper(self, data):
         normalized = np.clip(data[:, 0], -1, 1)
         means = self._bgm_transformer.means_.reshape([-1])
         stds = np.sqrt(self._bgm_transformer.covariances_).reshape([-1])
-        selected_component = data[:, 1].astype(int)  # maybe round instead?
+        selected_component = data[:, 1].round().astype(int)
+        selected_component = selected_component.clip(0, self.valid_component_indicator.sum() - 1)
         std_t = stds[self.valid_component_indicator][selected_component]
         mean_t = means[self.valid_component_indicator][selected_component]
         reversed_data = normalized * self.STD_MULTIPLIER * std_t + mean_t
@@ -546,8 +548,6 @@ def _reverse_transform(self, data):
 
         recovered_data = self._reverse_transform_helper(data)
         if self.null_transformer and self.null_transformer.models_missing_values():
-            data = np.stack([recovered_data, data[:, -1]], axis=1)  # noqa: PD013
-        else:
-            data = recovered_data
+            recovered_data = np.stack([recovered_data, data[:, -1]], axis=1)  # noqa: PD013
 
-        return super()._reverse_transform(data)
+        return super()._reverse_transform(recovered_data)

tests/unit/transformers/test_numerical.py

@@ -1213,7 +1213,6 @@ def test__fit(self, mock_bgm):
             n_components=10,
             weight_concentration_prior_type='dirichlet_process',
             weight_concentration_prior=0.001,
-            n_init=1,
             random_state=0
         )