removing minhash🧹

soda-inria · Nov 4, 2024 · 86c94b2 · 86c94b2
1 parent 93c77d2
commit 86c94b2
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 31 deletions.
diff --git a/carte_ai/src/carte_table_to_graph.py b/carte_ai/src/carte_table_to_graph.py
@@ -139,22 +139,6 @@ def fit(self, X, y=None):
 
         self.num_transformer_ = PowerTransformer().set_output(transform="pandas")
 
-        if self.lm_model == "minhash":
-            self.name_transformer = make_pipeline(
-                FeatureHasher(n_features=self.n_components, input_type="string"),
-                PowerTransformer(),
-            )
-            # Collect all unique names to fit the PowerTransformer
-            names_total = self.col_names
-            # Transform names using FeatureHasher
-            hashed_features = self.name_transformer.named_steps[
-                "featurehasher"
-            ].transform(names_total)
-            hashed_features_dense = hashed_features.toarray()
-            # Fit the PowerTransformer
-            self.name_transformer.named_steps["powertransformer"].fit(
-                hashed_features_dense
-            )
 
         # Ensure numerical columns exist before fitting the transformer
         if self.num_col_names:
@@ -238,9 +222,7 @@ def _load_lm_model(self):
             self.lm_model_ = fasttext.load_model(self.fasttext_model_path)
             if self.n_components != 300:
                 fasttext.util.reduce_model(self.lm_model_, self.n_components)
-        elif self.lm_model == "minhash":
-            # No need to load a model for FeatureHasher
-            pass
+
 
     def _transform_numerical(self, X):
         """
@@ -277,17 +259,6 @@ def _transform_names(self, names_total):
                 [self.lm_model_.get_sentence_vector(name) for name in names_total],
                 dtype=np.float32,
             )
-        elif self.lm_model == "minhash":
-            # Transform names using FeatureHasher
-            hashed_features = self.name_transformer.named_steps[
-                "featurehasher"
-            ].transform(names_total)
-            hashed_features_dense = hashed_features.toarray()
-            # Apply PowerTransformer
-            transformed_features = self.name_transformer.named_steps[
-                "powertransformer"
-            ].transform(hashed_features_dense)
-            return transformed_features.astype(np.float32)
 
     def _graph_construct(self, data_cat, data_num, name_attr_total, name_dict, y, idx):
         """

diff --git a/tests/tests_src/test_carte_table_to_graph.py b/tests/tests_src/test_carte_table_to_graph.py
@@ -62,7 +62,7 @@ def test_invalid_fasttext_model_path(dummy_data):
         transformer = Table2GraphTransformer(lm_model="fasttext")
         transformer.fit(dummy_data)
 
-@pytest.mark.parametrize("lm_model", ["fasttext", "minhash"])
+@pytest.mark.parametrize("lm_model", ["fasttext"])
 def test_table_to_graph_with_different_lm_models(dummy_data, lm_model):
     """Test Table2GraphTransformer with different language models."""
     transformer = Table2GraphTransformer(lm_model=lm_model, n_components=100, fasttext_model_path="path/to/fasttext.bin")