Add n_text_features and type_text features to Khiops sklearn supervised estimators

Author: popescu-v

khiops/sklearn/estimators.py

@@ -1199,6 +1199,8 @@ def __init__(
         self,
         n_features=100,
         n_trees=10,
+        n_text_features=10000,
+        type_text_features="words",
         specific_pairs=None,
         all_possible_pairs=True,
         construction_rules=None,
@@ -1213,6 +1215,8 @@ def __init__(
         )
         self.n_features = n_features
         self.n_trees = n_trees
+        self.n_text_features = n_text_features
+        self.type_text_features = type_text_features
         self.specific_pairs = specific_pairs
         self.all_possible_pairs = all_possible_pairs
         self.construction_rules = construction_rules
@@ -1280,6 +1284,20 @@ def _fit_check_params(self, ds, **kwargs):
             raise TypeError(type_error_message("n_trees", self.n_trees, int))
         if self.n_trees < 0:
             raise ValueError("'n_trees' must be positive")
+        if not isinstance(self.n_text_features, int):
+            raise TypeError(
+                type_error_message("n_text_features", self.n_text_features, int)
+            )
+        if self.n_text_features < 0:
+            raise ValueError("'n_text_features' must be positive")
+        if not isinstance(self.type_text_features, str):
+            raise TypeError(
+                type_error_message("type_text_features", self.type_text_features, str)
+            )
+        if self.type_text_features not in ("words", "ngrams", "tokens"):
+            raise ValueError(
+                "'type_text_features' must be among 'words', 'ngrams' or 'tokens'"
+            )
         if self.construction_rules is not None:
             if not is_list_like(self.construction_rules):
                 raise TypeError(
@@ -1373,6 +1391,8 @@ def _fit_prepare_training_function_inputs(self, ds, computation_dir):
         # Rename parameters to be compatible with khiops.core
         kwargs["max_constructed_variables"] = kwargs.pop("n_features")
         kwargs["max_trees"] = kwargs.pop("n_trees")
+        kwargs["max_text_features"] = kwargs.pop("n_text_features")
+        kwargs["text_features"] = kwargs.pop("type_text_features")
 
         # Add the additional_data_tables parameter
         kwargs["additional_data_tables"] = additional_data_tables
@@ -1548,6 +1568,8 @@ def __init__(
         self,
         n_features=100,
         n_trees=10,
+        n_text_features=10000,
+        type_text_features="words",
         n_selected_features=0,
         n_evaluated_features=0,
         specific_pairs=None,
@@ -1560,6 +1582,8 @@ def __init__(
         super().__init__(
             n_features=n_features,
             n_trees=n_trees,
+            n_text_features=n_text_features,
+            type_text_features=type_text_features,
             specific_pairs=specific_pairs,
             all_possible_pairs=all_possible_pairs,
             construction_rules=construction_rules,
@@ -1722,6 +1746,13 @@ class KhiopsClassifier(ClassifierMixin, KhiopsPredictor):
         combine other features, either native or constructed. These features usually
         improve the classifier's performance at the cost of interpretability of the
         model.
+    n_text_features : int, default 10000
+        Maximum number of text features to construct.
+    type_text_features : str, default "words"
+        Type of the text features to construct. Can be either one of:
+        - "words": sequences of non-space characters
+        - "ngrams": sequences of bytes
+        - "tokens": user-defined
     n_selected_features : int, default 0
         Maximum number of features to be selected in the SNB predictor. If equal to
         0 it selects all the features kept in the training.
@@ -1813,6 +1844,8 @@ def __init__(
         n_features=100,
         n_pairs=0,
         n_trees=10,
+        n_text_features=10000,
+        type_text_features="words",
         n_selected_features=0,
         n_evaluated_features=0,
         specific_pairs=None,
@@ -1826,6 +1859,8 @@ def __init__(
         super().__init__(
             n_features=n_features,
             n_trees=n_trees,
+            n_text_features=n_text_features,
+            type_text_features=type_text_features,
             n_selected_features=n_selected_features,
             n_evaluated_features=n_evaluated_features,
             construction_rules=construction_rules,
@@ -2217,6 +2252,8 @@ def __init__(
         self,
         n_features=100,
         n_trees=0,
+        n_text_features=10000,
+        type_text_features="words",
         n_selected_features=0,
         n_evaluated_features=0,
         construction_rules=None,
@@ -2227,6 +2264,8 @@ def __init__(
         super().__init__(
             n_features=n_features,
             n_trees=n_trees,
+            n_text_features=n_text_features,
+            type_text_features=type_text_features,
             n_selected_features=n_selected_features,
             n_evaluated_features=n_evaluated_features,
             construction_rules=construction_rules,
@@ -2376,6 +2415,13 @@ class KhiopsEncoder(TransformerMixin, KhiopsSupervisedEstimator):
         Maximum number of decision tree features to construct. The constructed trees
         combine other features, either native or constructed. These features usually
         improve a predictor's performance at the cost of interpretability of the model.
+    n_text_features : int, default 10000
+        Maximum number of text features to construct.
+    type_text_features : str, default "words"
+        Type of the text features to construct. Can be either one of:
+        - "words": sequences of non-space characters
+        - "ngrams": sequences of bytes
+        - "tokens": user-defined
     specific_pairs : list of tuple, optional
         User-specified pairs as a list of 2-tuples of feature names. If a given tuple
         contains only one non-empty feature name, then it generates all the pairs
@@ -2469,6 +2515,8 @@ def __init__(
         n_features=100,
         n_pairs=0,
         n_trees=0,
+        n_text_features=10000,
+        type_text_features="words",
         specific_pairs=None,
         all_possible_pairs=True,
         construction_rules=None,
@@ -2485,6 +2533,8 @@ def __init__(
         super().__init__(
             n_features=n_features,
             n_trees=n_trees,
+            n_text_features=n_text_features,
+            type_text_features=type_text_features,
             construction_rules=construction_rules,
             verbose=verbose,
             output_dir=output_dir,

tests/test_estimator_attributes.py

@@ -187,7 +187,7 @@ def test_classifier_attributes_multitable(self):
          by Khiops post training.
         """
         X, y = self._create_multitable_input()
-        khc_accidents = KhiopsClassifier(n_trees=0, n_pairs=10)
+        khc_accidents = KhiopsClassifier(n_trees=0, n_text_features=0, n_pairs=10)
         khc_accidents.fit(X, y)
         self.assert_attribute_values_ok(khc_accidents, X["main_table"][0], y)
         self.assertTrue(khc_accidents.is_multitable_model_)
@@ -203,7 +203,7 @@ def test_regressor_attributes_monotable(self):
         adult_df = pd.read_csv(adult_dataset_path, sep="\t").sample(750)
         X = adult_df.drop("age", axis=1)
         y = adult_df["age"]
-        khr_adult = KhiopsRegressor(n_trees=0)
+        khr_adult = KhiopsRegressor(n_trees=0, n_text_features=0)
         with warnings.catch_warnings():
             warnings.filterwarnings(
                 action="ignore",
@@ -225,7 +225,7 @@ def test_regressor_attributes_multitable(self):
         X, _ = self._create_multitable_input(750)
         y = X["main_table"][0]["Commune"]
         X["main_table"][0].drop("Commune", axis=1, inplace=True)
-        khr_accidents = KhiopsRegressor(n_trees=0)
+        khr_accidents = KhiopsRegressor(n_trees=0, n_text_features=0)
         with warnings.catch_warnings():
             warnings.filterwarnings(
                 action="ignore",
@@ -262,7 +262,7 @@ def test_encoder_attributes_multitable(self):
          by Khiops post training.
         """
         X, y = self._create_multitable_input()
-        khe_accidents = KhiopsEncoder(n_trees=5)
+        khe_accidents = KhiopsEncoder(n_trees=5, n_text_features=5000)
         khe_accidents.fit(X, y)
 
         self.assert_attribute_values_ok(khe_accidents, X, None)

tests/test_sklearn.py

@@ -749,6 +749,8 @@ def setUpClass(cls):
                                 "header_line": True,
                                 "max_pairs": 1,
                                 "max_trees": 5,
+                                "max_text_features": 300000,
+                                "text_features": "ngrams",
                                 "max_selected_variables": 1,
                                 "max_evaluated_variables": 3,
                                 "specific_pairs": [("age", "race")],
@@ -777,6 +779,8 @@ def setUpClass(cls):
                                 "detect_format": False,
                                 "header_line": True,
                                 "max_trees": 0,
+                                "max_text_features": 300000,
+                                "text_features": "ngrams",
                                 "max_selected_variables": 1,
                                 "max_evaluated_variables": 3,
                                 "construction_rules": ["TableMode", "TableSelection"],
@@ -803,6 +807,8 @@ def setUpClass(cls):
                                 "header_line": True,
                                 "max_pairs": 1,
                                 "max_trees": 5,
+                                "max_text_features": 300000,
+                                "text_features": "ngrams",
                                 "specific_pairs": [("age", "race")],
                                 "all_possible_pairs": False,
                                 "construction_rules": ["TableMode", "TableSelection"],
@@ -841,6 +847,8 @@ def setUpClass(cls):
                                 "max_constructed_variables": 10,
                                 "max_pairs": 1,
                                 "max_trees": 5,
+                                "max_text_features": 300000,
+                                "text_features": "ngrams",
                                 "max_selected_variables": 1,
                                 "max_evaluated_variables": 3,
                                 "specific_pairs": [],
@@ -870,6 +878,8 @@ def setUpClass(cls):
                                 "header_line": True,
                                 "max_constructed_variables": 10,
                                 "max_trees": 0,
+                                "max_text_features": 300000,
+                                "text_features": "ngrams",
                                 "max_selected_variables": 1,
                                 "max_evaluated_variables": 3,
                                 "construction_rules": ["TableMode", "TableSelection"],
@@ -897,6 +907,8 @@ def setUpClass(cls):
                                 "max_constructed_variables": 10,
                                 "max_pairs": 1,
                                 "max_trees": 5,
+                                "max_text_features": 300000,
+                                "text_features": "ngrams",
                                 "specific_pairs": [],
                                 "all_possible_pairs": False,
                                 "construction_rules": ["TableMode", "TableSelection"],
@@ -1410,6 +1422,8 @@ def test_parameter_transfer_classifier_fit_from_monotable_dataframe(self):
             extra_estimator_kwargs={
                 "n_pairs": 1,
                 "n_trees": 5,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "n_selected_features": 1,
                 "n_evaluated_features": 3,
                 "specific_pairs": [("age", "race")],
@@ -1431,6 +1445,8 @@ def test_parameter_transfer_classifier_fit_from_monotable_dataframe_with_df_y(
             extra_estimator_kwargs={
                 "n_pairs": 1,
                 "n_trees": 5,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "n_selected_features": 1,
                 "n_evaluated_features": 3,
                 "specific_pairs": [("age", "race")],
@@ -1451,6 +1467,8 @@ def test_parameter_transfer_classifier_fit_from_multitable_dataframe(self):
                 "n_features": 10,
                 "n_pairs": 1,
                 "n_trees": 5,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "n_selected_features": 1,
                 "n_evaluated_features": 3,
                 "specific_pairs": [],
@@ -1488,6 +1506,8 @@ def test_parameter_transfer_encoder_fit_from_monotable_dataframe(self):
             extra_estimator_kwargs={
                 "n_pairs": 1,
                 "n_trees": 5,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "specific_pairs": [("age", "race")],
                 "all_possible_pairs": False,
                 "construction_rules": ["TableMode", "TableSelection"],
@@ -1512,6 +1532,8 @@ def test_parameter_transfer_encoder_fit_from_monotable_dataframe_with_df_y(
             extra_estimator_kwargs={
                 "n_pairs": 1,
                 "n_trees": 5,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "specific_pairs": [("age", "race")],
                 "all_possible_pairs": False,
                 "construction_rules": ["TableMode", "TableSelection"],
@@ -1535,6 +1557,8 @@ def test_parameter_transfer_encoder_fit_from_multitable_dataframe(self):
                 "n_features": 10,
                 "n_pairs": 1,
                 "n_trees": 5,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "specific_pairs": [],
                 "all_possible_pairs": False,
                 "construction_rules": ["TableMode", "TableSelection"],
@@ -1575,6 +1599,8 @@ def test_parameter_transfer_regressor_fit_from_monotable_dataframe(self):
             extra_estimator_kwargs={
                 "n_selected_features": 1,
                 "n_evaluated_features": 3,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "construction_rules": ["TableMode", "TableSelection"],
             },
         )
@@ -1591,6 +1617,8 @@ def test_parameter_transfer_regressor_fit_from_monotable_dataframe_with_df_y(
             extra_estimator_kwargs={
                 "n_selected_features": 1,
                 "n_evaluated_features": 3,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "construction_rules": ["TableMode", "TableSelection"],
             },
         )
@@ -1605,6 +1633,8 @@ def test_parameter_transfer_regressor_fit_from_multitable_dataframe(self):
             extra_estimator_kwargs={
                 "n_features": 10,
                 "n_trees": 0,
+                "n_text_features": 300000,
+                "type_text_features": "ngrams",
                 "n_selected_features": 1,
                 "n_evaluated_features": 3,
                 "construction_rules": ["TableMode", "TableSelection"],
@@ -1693,6 +1723,7 @@ def test_sklearn_check_estimator(self):
         # Set the estimators to test
         # Notes:
         # - We use n_trees=0 so the tests execute faster
+        # - We use n_text_features=0 so the tests execute faster
         # - We omit KhiopsCoclustering because he needs special inputs to work well
         #   and sklearn's check_estimator method does not accept them.
         # - KhiopsEncoder:
@@ -1701,10 +1732,11 @@ def test_sklearn_check_estimator(self):
         #   - We set it with informative_features_only=False so it always have output
         #   columns (sklearn estimator checks expect non-empty encoders)
         khiops_estimators = [
-            KhiopsClassifier(n_trees=0),
-            KhiopsRegressor(n_trees=0),
+            KhiopsClassifier(n_trees=0, n_text_features=0),
+            KhiopsRegressor(n_trees=0, n_text_features=0),
             KhiopsEncoder(
                 n_trees=0,
+                n_text_features=0,
                 informative_features_only=False,
                 transform_type_numerical="0-1_normalization",
             ),

tests/test_sklearn_output_types.py

@@ -66,7 +66,7 @@ def test_classifier_output_types(self):
                 "iris_sec": (raw_X_sec_mt, ["Id"]),
             },
         }
-        khc = KhiopsClassifier(n_trees=0)
+        khc = KhiopsClassifier(n_trees=0, n_text_features=0)
         khc.fit(X, y)
         y_pred = khc.predict(X)
         khc.fit(X_mt, y)
@@ -185,7 +185,7 @@ def test_classifier_output_types(self):
                     estimator=KhiopsClassifier.__name__,
                 ):
                     # Train the classifier
-                    khc = KhiopsClassifier(n_trees=0)
+                    khc = KhiopsClassifier(n_trees=0, n_text_features=0)
                     khc.fit(X, y)
 
                     # Check the expected classes