autogluon.features 1.5.1b20260107__tar.gz → 1.5.1b20260109__tar.gz

{autogluon_features-1.5.1b20260107/src/autogluon.features.egg-info → autogluon_features-1.5.1b20260109}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: autogluon.features
-Version: 1.5.1b20260107
+Version: 1.5.1b20260109
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community
@@ -38,7 +38,7 @@ License-File: NOTICE
 Requires-Dist: numpy<2.4.0,>=1.25.0
 Requires-Dist: pandas<2.4.0,>=2.0.0
 Requires-Dist: scikit-learn<1.8.0,>=1.4.0
-Requires-Dist: autogluon.common==1.5.1b20260107
+Requires-Dist: autogluon.common==1.5.1b20260109
 Dynamic: author
 Dynamic: classifier
 Dynamic: description

{autogluon_features-1.5.1b20260107 → autogluon_features-1.5.1b20260109}/src/autogluon/features/generators/abstract.py

@@ -558,6 +558,27 @@ class AbstractFeatureGenerator:
     def get_default_infer_features_in_args() -> dict:
         raise NotImplementedError
 
+    @staticmethod
+    def get_infer_features_in_args_to_drop() -> dict:
+        """Return a dict of kwargs for FeatureMetadata.get_features().
+
+        This allows to specify which features should be dropped after running this
+        feature generator in a feature generator group.
+
+         For example, assume you are using a feature generator to apply PCA to all
+         features of special type S_TEXT_EMBEDDING, then this function could return:
+            {
+                "invalid_special_types": [S_TEXT_EMBEDDING]
+            }
+        to inform the user that all S_TEXT_EMBEDDING features that are captured by PCA
+        should be dropped from the output of the feature generator group.
+        """
+        return {}
+
+    def estimate_output_feature_metadata(self, feature_metadata_in: FeatureMetadata, **kwargs) -> FeatureMetadata:
+        """Return an estimated representation of the feature metadata after fit_transform."""
+        raise NotImplementedError("This method is not implemented for this generator.")
+
     def _fit_generators(
         self, X, y, feature_metadata, generators: list["AbstractFeatureGenerator"], **kwargs
     ) -> (DataFrame, FeatureMetadata, list):
@@ -918,3 +939,23 @@ class AbstractFeatureGenerator:
                 more_tags = base_class._more_tags(self)
                 collected_tags.update(more_tags)
         return collected_tags
+
+
+# FIXME: this logic still needs more work to become general purpose.
+#   - Needs to make it work for multiple feature generator groups
+#   - Need to support for all possible feature generators
+def estimate_feature_metadata_after_generators(
+    *, feature_generators: list[list[AbstractFeatureGenerator]] | None, feature_metadata_in: FeatureMetadata, **kwargs
+) -> FeatureMetadata:
+    """Estimate the feature metadata after applying a set of feature generators."""
+    feature_metadata = copy.deepcopy(feature_metadata_in)
+    if feature_generators is not None:
+        for fg_group in feature_generators:
+            feature_metadatas = [
+                fg.estimate_output_feature_metadata(feature_metadata_in=feature_metadata, **kwargs) for fg in fg_group
+            ]
+            feature_metadata = FeatureMetadata.join_metadatas(
+                feature_metadatas,
+                shared_raw_features="error",
+            )
+    return feature_metadata

{autogluon_features-1.5.1b20260107 → autogluon_features-1.5.1b20260109}/src/autogluon/features/generators/auto_ml_pipeline.py

@@ -1,6 +1,17 @@
 import logging
+from enum import Enum
 
-from autogluon.common.features.types import R_FLOAT, R_INT, R_OBJECT, S_IMAGE_BYTEARRAY, S_IMAGE_PATH, S_TEXT
+from sklearn.feature_extraction.text import CountVectorizer
+
+from autogluon.common.features.types import (
+    R_FLOAT,
+    R_INT,
+    R_OBJECT,
+    S_IMAGE_BYTEARRAY,
+    S_IMAGE_PATH,
+    S_TEXT,
+)
+from autogluon.features.generators.abstract import AbstractFeatureGenerator
 
 from .category import CategoryFeatureGenerator
 from .datetime import DatetimeFeatureGenerator
@@ -17,8 +28,8 @@ logger = logging.getLogger(__name__)
 # TODO: write out in English the full set of transformations that are applied (and eventually host page on website).
 #  Also explicitly write out all of the feature-generator "hyperparameters" that might affect the results from the AutoML FeatureGenerator
 class AutoMLPipelineFeatureGenerator(PipelineFeatureGenerator):
-    """
-    Pipeline feature generator with simplified arguments to handle most Tabular data including text and dates adequately.
+    """Pipeline feature generator with simplified arguments to handle most Tabular data including text and dates adequately.
+
     This is the default feature generation pipeline used by AutoGluon when unspecified.
     For more customization options, refer to :class:`PipelineFeatureGenerator` and :class:`BulkFeatureGenerator`.
 
@@ -56,6 +67,17 @@ class AutoMLPipelineFeatureGenerator(PipelineFeatureGenerator):
     vectorizer : :class:`sklearn.feature_extraction.text.CountVectorizer`, default CountVectorizer(min_df=30, ngram_range=(1, 3), max_features=10000, dtype=np.uint8)  # noqa
         sklearn CountVectorizer object to use in :class:`TextNgramFeatureGenerator`.
         Only used if `enable_text_ngram_features=True`.
+    text_ngram_params : dict, default None
+        Parameters besides vectorizer passed to the :class:`TextNgramFeatureGenerator`.
+    custom_feature_generators : list of :class:`AbstractFeatureGenerator`, default None
+        Lists of custom feature generators. This list is inserted in the first generator
+        step that is getting the original X data (i.e. before any other feature
+        generators are applied) as input.
+        Note, there might be an overlap of custom feature generators with the default
+        feature generators used in AutoMLPipelineFeatureGenerator. It is the user's
+        responsibility to avoid unwanted overlap and disable default generators if
+        needed.
+        If None, no custom feature generators are added.
     **kwargs :
         Refer to :class:`AbstractFeatureGenerator` documentation for details on valid key word arguments.
 
@@ -80,15 +102,16 @@ class AutoMLPipelineFeatureGenerator(PipelineFeatureGenerator):
 
     def __init__(
         self,
-        enable_numeric_features=True,
-        enable_categorical_features=True,
-        enable_datetime_features=True,
-        enable_text_special_features=True,
-        enable_text_ngram_features=True,
-        enable_raw_text_features=False,
-        enable_vision_features=True,
-        vectorizer=None,
-        text_ngram_params=None,
+        enable_numeric_features: bool = True,
+        enable_categorical_features: bool = True,
+        enable_datetime_features: bool = True,
+        enable_text_special_features: bool = True,
+        enable_text_ngram_features: bool = True,
+        enable_raw_text_features: bool = False,
+        enable_vision_features: bool = True,
+        vectorizer: CountVectorizer | None = None,
+        text_ngram_params: dict | None = None,
+        custom_feature_generators: list[AbstractFeatureGenerator] | None = None,
         **kwargs,
     ):
         if "generators" in kwargs:
@@ -111,12 +134,15 @@ class AutoMLPipelineFeatureGenerator(PipelineFeatureGenerator):
         self.enable_raw_text_features = enable_raw_text_features
         self.enable_vision_features = enable_vision_features
         self.text_ngram_params = text_ngram_params if text_ngram_params else {}
+        self.custom_feature_generators = custom_feature_generators
 
         generators = self._get_default_generators(vectorizer=vectorizer)
         super().__init__(generators=generators, **kwargs)
 
+    # TODO: switch to / add skrub's String or Text encoders
     def _get_default_generators(self, vectorizer=None):
         generator_group = []
+
         if self.enable_numeric_features:
             generator_group.append(
                 IdentityFeatureGenerator(infer_features_in_args=dict(valid_raw_types=[R_INT, R_FLOAT]))
@@ -125,7 +151,8 @@ class AutoMLPipelineFeatureGenerator(PipelineFeatureGenerator):
             generator_group.append(
                 IdentityFeatureGenerator(
                     infer_features_in_args=dict(
-                        required_special_types=[S_TEXT], invalid_special_types=[S_IMAGE_PATH, S_IMAGE_BYTEARRAY]
+                        required_special_types=[S_TEXT],
+                        invalid_special_types=[S_IMAGE_PATH, S_IMAGE_BYTEARRAY],
                     ),
                     name_suffix="_raw_text",
                 )
@@ -157,6 +184,13 @@ class AutoMLPipelineFeatureGenerator(PipelineFeatureGenerator):
                     )
                 )
             )
+
+        if self.custom_feature_generators is not None:
+            generator_group = [
+                *generator_group,
+                *self.custom_feature_generators,
+            ]
+
         generators = [generator_group]
         return generators
 
@@ -167,5 +201,7 @@ class AutoMLPipelineFeatureGenerator(PipelineFeatureGenerator):
 class AutoMLInterpretablePipelineFeatureGenerator(AutoMLPipelineFeatureGenerator):
     def _get_category_feature_generator(self):
         return CategoryFeatureGenerator(
-            minimize_memory=False, maximum_num_cat=10, post_generators=[OneHotEncoderFeatureGenerator()]
+            minimize_memory=False,
+            maximum_num_cat=10,
+            post_generators=[OneHotEncoderFeatureGenerator()],
         )

{autogluon_features-1.5.1b20260107 → autogluon_features-1.5.1b20260109}/src/autogluon/features/generators/bulk.py

@@ -93,7 +93,7 @@ class BulkFeatureGenerator(AbstractFeatureGenerator):
     def __init__(
         self,
         generators: list[list[AbstractFeatureGenerator | list]],
-        pre_generators: list[AbstractFeatureGenerator] = None,
+        pre_generators: list[AbstractFeatureGenerator | List[AbstractFeatureGenerator]] = None,
         remove_unused_features: bool | str = True,
         **kwargs,
     ):
@@ -143,10 +143,15 @@ class BulkFeatureGenerator(AbstractFeatureGenerator):
 
             pre_generators = [AsTypeFeatureGenerator()] + pre_generators
             self.pre_enforce_types = False
-        pre_generators = [[pre_generator] for pre_generator in pre_generators]
+        pre_generators = [
+            pre_generator if isinstance(pre_generator, list) else [pre_generator] for pre_generator in pre_generators
+        ]
 
         if self._post_generators is not None:
-            post_generators = [[post_generator] for post_generator in self._post_generators]
+            post_generators = [
+                post_generator if isinstance(post_generator, list) else [post_generator]
+                for post_generator in self._post_generators
+            ]
             self._post_generators = []
         else:
             post_generators = []

{autogluon_features-1.5.1b20260107 → autogluon_features-1.5.1b20260109}/src/autogluon/features/generators/identity.py

@@ -2,6 +2,8 @@ import logging
 
 from pandas import DataFrame
 
+from autogluon.common.features.feature_metadata import FeatureMetadata
+
 from .abstract import AbstractFeatureGenerator
 
 logger = logging.getLogger(__name__)
@@ -23,3 +25,7 @@ class IdentityFeatureGenerator(AbstractFeatureGenerator):
 
     def _more_tags(self):
         return {"feature_interactions": False}
+
+    def estimate_output_feature_metadata(self, feature_metadata_in: FeatureMetadata, **kwargs) -> FeatureMetadata:
+        features_to_remove = feature_metadata_in.get_features(**self._infer_features_in_args)
+        return feature_metadata_in.keep_features(features_to_remove, inplace=False)

{autogluon_features-1.5.1b20260107 → autogluon_features-1.5.1b20260109}/src/autogluon/features/version.py

@@ -1,4 +1,4 @@
 """This is the autogluon version file."""
 
-__version__ = "1.5.1b20260107"
+__version__ = "1.5.1b20260109"
 __lite__ = False

{autogluon_features-1.5.1b20260107 → autogluon_features-1.5.1b20260109/src/autogluon.features.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: autogluon.features
-Version: 1.5.1b20260107
+Version: 1.5.1b20260109
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community
@@ -38,7 +38,7 @@ License-File: NOTICE
 Requires-Dist: numpy<2.4.0,>=1.25.0
 Requires-Dist: pandas<2.4.0,>=2.0.0
 Requires-Dist: scikit-learn<1.8.0,>=1.4.0
-Requires-Dist: autogluon.common==1.5.1b20260107
+Requires-Dist: autogluon.common==1.5.1b20260109
 Dynamic: author
 Dynamic: classifier
 Dynamic: description

{autogluon_features-1.5.1b20260107 → autogluon_features-1.5.1b20260109}/src/autogluon.features.egg-info/requires.txt

@@ -1,4 +1,4 @@
 numpy<2.4.0,>=1.25.0
 pandas<2.4.0,>=2.0.0
 scikit-learn<1.8.0,>=1.4.0
-autogluon.common==1.5.1b20260107
+autogluon.common==1.5.1b20260109