feat: add ColumnTransformer save/load (#541)

GarrettWu · web-flow · commit 9d8cf6792a8d · 2024-03-29T14:12:03.000-07:00
diff --git a/bigframes/ml/compose.py b/bigframes/ml/compose.py
@@ -18,16 +18,21 @@
 
 from __future__ import annotations
 
+import re
+import types
 import typing
-from typing import List, Optional, Tuple, Union
+from typing import cast, List, Optional, Tuple, Union
 
 import bigframes_vendored.sklearn.compose._column_transformer
+from google.cloud import bigquery
 
+import bigframes
+from bigframes import constants
 from bigframes.core import log_adapter
 from bigframes.ml import base, core, globals, preprocessing, utils
 import bigframes.pandas as bpd
 
-CompilablePreprocessorType = Union[
+_PREPROCESSING_TYPES = Union[
     preprocessing.OneHotEncoder,
     preprocessing.StandardScaler,
     preprocessing.MaxAbsScaler,
@@ -36,6 +41,17 @@
     preprocessing.LabelEncoder,
 ]
 
+_BQML_TRANSFROM_TYPE_MAPPING = types.MappingProxyType(
+    {
+        "ML.STANDARD_SCALER": preprocessing.StandardScaler,
+        "ML.ONE_HOT_ENCODER": preprocessing.OneHotEncoder,
+        "ML.MAX_ABS_SCALER": preprocessing.MaxAbsScaler,
+        "ML.MIN_MAX_SCALER": preprocessing.MinMaxScaler,
+        "ML.BUCKETIZE": preprocessing.KBinsDiscretizer,
+        "ML.LABEL_ENCODER": preprocessing.LabelEncoder,
+    }
+)
+
 
 @log_adapter.class_logger
 class ColumnTransformer(
@@ -51,7 +67,7 @@ def __init__(
         transformers: List[
             Tuple[
                 str,
-                CompilablePreprocessorType,
+                _PREPROCESSING_TYPES,
                 Union[str, List[str]],
             ]
         ],
@@ -66,12 +82,12 @@ def __init__(
     @property
     def transformers_(
         self,
-    ) -> List[Tuple[str, CompilablePreprocessorType, str,]]:
+    ) -> List[Tuple[str, _PREPROCESSING_TYPES, str,]]:
         """The collection of transformers as tuples of (name, transformer, column)."""
         result: List[
             Tuple[
                 str,
-                CompilablePreprocessorType,
+                _PREPROCESSING_TYPES,
                 str,
             ]
         ] = []
@@ -89,6 +105,96 @@ def transformers_(
 
         return result
 
+    @classmethod
+    def _from_bq(
+        cls, session: bigframes.Session, model: bigquery.Model
+    ) -> ColumnTransformer:
+        col_transformer = cls._extract_from_bq_model(model)
+        col_transformer._bqml_model = core.BqmlModel(session, model)
+
+        return col_transformer
+
+    @classmethod
+    def _extract_from_bq_model(
+        cls,
+        bq_model: bigquery.Model,
+    ) -> ColumnTransformer:
+        """Extract transformers as ColumnTransformer obj from a BQ Model. Keep the _bqml_model field as None."""
+        assert "transformColumns" in bq_model._properties
+
+        transformers: List[
+            Tuple[
+                str,
+                _PREPROCESSING_TYPES,
+                Union[str, List[str]],
+            ]
+        ] = []
+
+        def camel_to_snake(name):
+            name = re.sub("(.)([A-Z][a-z]+)", r"\1_\2", name)
+            return re.sub("([a-z0-9])([A-Z])", r"\1_\2", name).lower()
+
+        for transform_col in bq_model._properties["transformColumns"]:
+            # pass the columns that are not transformed
+            if "transformSql" not in transform_col:
+                continue
+            transform_sql: str = cast(dict, transform_col)["transformSql"]
+            if not transform_sql.startswith("ML."):
+                continue
+
+            found_transformer = False
+            for prefix in _BQML_TRANSFROM_TYPE_MAPPING:
+                if transform_sql.startswith(prefix):
+                    transformer_cls = _BQML_TRANSFROM_TYPE_MAPPING[prefix]
+                    transformers.append(
+                        (
+                            camel_to_snake(transformer_cls.__name__),
+                            *transformer_cls._parse_from_sql(transform_sql),  # type: ignore
+                        )
+                    )
+
+                    found_transformer = True
+                    break
+            if not found_transformer:
+                raise NotImplementedError(
+                    f"Unsupported transformer type. {constants.FEEDBACK_LINK}"
+                )
+
+        return cls(transformers=transformers)
+
+    def _merge(
+        self, bq_model: bigquery.Model
+    ) -> Union[
+        ColumnTransformer,
+        preprocessing.StandardScaler,
+        preprocessing.OneHotEncoder,
+        preprocessing.MaxAbsScaler,
+        preprocessing.MinMaxScaler,
+        preprocessing.KBinsDiscretizer,
+        preprocessing.LabelEncoder,
+    ]:
+        """Try to merge the column transformer to a simple transformer. Depends on all the columns in bq_model are transformed with the same transformer."""
+        transformers = self.transformers_
+
+        assert len(transformers) > 0
+        _, transformer_0, column_0 = transformers[0]
+        columns = [column_0]
+        for _, transformer, column in transformers[1:]:
+            # all transformers are the same
+            if transformer != transformer_0:
+                return self
+            columns.append(column)
+        # all feature columns are transformed
+        if sorted(
+            [
+                cast(str, feature_column.name)
+                for feature_column in bq_model.feature_columns
+            ]
+        ) == sorted(columns):
+            return transformer_0
+
+        return self
+
     def _compile_to_sql(
         self,
         columns: List[str],
@@ -143,3 +249,20 @@ def transform(self, X: Union[bpd.DataFrame, bpd.Series]) -> bpd.DataFrame:
             bpd.DataFrame,
             df[self._output_names],
         )
+
+    def to_gbq(self, model_name: str, replace: bool = False) -> ColumnTransformer:
+        """Save the transformer as a BigQuery model.
+
+        Args:
+            model_name (str):
+                the name of the model.
+            replace (bool, default False):
+                whether to replace if the model already exists. Default to False.
+
+        Returns:
+            ColumnTransformer: saved model."""
+        if not self._bqml_model:
+            raise RuntimeError("A transformer must be fitted before it can be saved")
+
+        new_model = self._bqml_model.copy(model_name, replace)
+        return new_model.session.read_gbq_model(model_name)
diff --git a/bigframes/ml/loader.py b/bigframes/ml/loader.py
@@ -23,6 +23,7 @@
 import bigframes.constants as constants
 from bigframes.ml import (
     cluster,
+    compose,
     decomposition,
     ensemble,
     forecasting,
@@ -79,6 +80,7 @@ def from_bq(
     llm.PaLM2TextGenerator,
     llm.PaLM2TextEmbeddingGenerator,
     pipeline.Pipeline,
+    compose.ColumnTransformer,
 ]:
     """Load a BQML model to BigQuery DataFrames ML.
 
@@ -89,22 +91,32 @@ def from_bq(
     Returns:
         A BigQuery DataFrames ML model object.
     """
+    # TODO(garrettwu): the entire condition only to TRANSFORM_ONLY when b/331679273 is fixed.
+    if (
+        bq_model.model_type == "TRANSFORM_ONLY"
+        or bq_model.model_type == "MODEL_TYPE_UNSPECIFIED"
+        and "transformColumns" in bq_model._properties
+        and not _is_bq_model_remote(bq_model)
+    ):
+        return _transformer_from_bq(session, bq_model)
+
     if _is_bq_model_pipeline(bq_model):
         return pipeline.Pipeline._from_bq(session, bq_model)
 
     return _model_from_bq(session, bq_model)
 
 
+def _transformer_from_bq(session: bigframes.Session, bq_model: bigquery.Model):
+    # TODO(garrettwu): add other transformers
+    return compose.ColumnTransformer._from_bq(session, bq_model)
+
+
 def _model_from_bq(session: bigframes.Session, bq_model: bigquery.Model):
     if bq_model.model_type in _BQML_MODEL_TYPE_MAPPING:
         return _BQML_MODEL_TYPE_MAPPING[bq_model.model_type]._from_bq(  # type: ignore
             session=session, model=bq_model
         )
-    if (
-        bq_model.model_type == "MODEL_TYPE_UNSPECIFIED"
-        and "remoteModelInfo" in bq_model._properties
-        and "endpoint" in bq_model._properties["remoteModelInfo"]
-    ):
+    if _is_bq_model_remote(bq_model):
         # Parse the remote model endpoint
         bqml_endpoint = bq_model._properties["remoteModelInfo"]["endpoint"]
         model_endpoint = bqml_endpoint.split("/")[-1]
@@ -121,3 +133,11 @@ def _model_from_bq(session: bigframes.Session, bq_model: bigquery.Model):
 
 def _is_bq_model_pipeline(bq_model: bigquery.Model) -> bool:
     return "transformColumns" in bq_model._properties
+
+
+def _is_bq_model_remote(bq_model: bigquery.Model) -> bool:
+    return (
+        bq_model.model_type == "MODEL_TYPE_UNSPECIFIED"
+        and "remoteModelInfo" in bq_model._properties
+        and "endpoint" in bq_model._properties["remoteModelInfo"]
+    )
diff --git a/bigframes/ml/pipeline.py b/bigframes/ml/pipeline.py
@@ -18,7 +18,7 @@
 
 from __future__ import annotations
 
-from typing import cast, List, Optional, Tuple, Union
+from typing import List, Optional, Tuple, Union
 
 import bigframes_vendored.sklearn.pipeline
 from google.cloud import bigquery
@@ -83,8 +83,8 @@ def __init__(self, steps: List[Tuple[str, base.BaseEstimator]]):
 
     @classmethod
     def _from_bq(cls, session: bigframes.Session, bq_model: bigquery.Model) -> Pipeline:
-        col_transformer = _extract_as_column_transformer(bq_model)
-        transform = _merge_column_transformer(bq_model, col_transformer)
+        col_transformer = compose.ColumnTransformer._extract_from_bq_model(bq_model)
+        transform = col_transformer._merge(bq_model)
 
         estimator = loader._model_from_bq(session, bq_model)
         return cls([("transform", transform), ("estimator", estimator)])
@@ -138,110 +138,3 @@ def to_gbq(self, model_name: str, replace: bool = False) -> Pipeline:
         new_model = self._estimator._bqml_model.copy(model_name, replace)
 
         return new_model.session.read_gbq_model(model_name)
-
-
-def _extract_as_column_transformer(
-    bq_model: bigquery.Model,
-) -> compose.ColumnTransformer:
-    """Extract transformers as ColumnTransformer obj from a BQ Model."""
-    assert "transformColumns" in bq_model._properties
-
-    transformers: List[
-        Tuple[
-            str,
-            Union[
-                preprocessing.OneHotEncoder,
-                preprocessing.StandardScaler,
-                preprocessing.MaxAbsScaler,
-                preprocessing.MinMaxScaler,
-                preprocessing.KBinsDiscretizer,
-                preprocessing.LabelEncoder,
-            ],
-            Union[str, List[str]],
-        ]
-    ] = []
-    for transform_col in bq_model._properties["transformColumns"]:
-        # pass the columns that are not transformed
-        if "transformSql" not in transform_col:
-            continue
-
-        transform_sql: str = cast(dict, transform_col)["transformSql"]
-        if transform_sql.startswith("ML.STANDARD_SCALER"):
-            transformers.append(
-                (
-                    "standard_scaler",
-                    *preprocessing.StandardScaler._parse_from_sql(transform_sql),
-                )
-            )
-        elif transform_sql.startswith("ML.ONE_HOT_ENCODER"):
-            transformers.append(
-                (
-                    "ont_hot_encoder",
-                    *preprocessing.OneHotEncoder._parse_from_sql(transform_sql),
-                )
-            )
-        elif transform_sql.startswith("ML.MAX_ABS_SCALER"):
-            transformers.append(
-                (
-                    "max_abs_scaler",
-                    *preprocessing.MaxAbsScaler._parse_from_sql(transform_sql),
-                )
-            )
-        elif transform_sql.startswith("ML.MIN_MAX_SCALER"):
-            transformers.append(
-                (
-                    "min_max_scaler",
-                    *preprocessing.MinMaxScaler._parse_from_sql(transform_sql),
-                )
-            )
-        elif transform_sql.startswith("ML.BUCKETIZE"):
-            transformers.append(
-                (
-                    "k_bins_discretizer",
-                    *preprocessing.KBinsDiscretizer._parse_from_sql(transform_sql),
-                )
-            )
-        elif transform_sql.startswith("ML.LABEL_ENCODER"):
-            transformers.append(
-                (
-                    "label_encoder",
-                    *preprocessing.LabelEncoder._parse_from_sql(transform_sql),
-                )
-            )
-        else:
-            raise NotImplementedError(
-                f"Unsupported transformer type. {constants.FEEDBACK_LINK}"
-            )
-
-    return compose.ColumnTransformer(transformers=transformers)
-
-
-def _merge_column_transformer(
-    bq_model: bigquery.Model, column_transformer: compose.ColumnTransformer
-) -> Union[
-    compose.ColumnTransformer,
-    preprocessing.StandardScaler,
-    preprocessing.OneHotEncoder,
-    preprocessing.MaxAbsScaler,
-    preprocessing.MinMaxScaler,
-    preprocessing.KBinsDiscretizer,
-    preprocessing.LabelEncoder,
-]:
-    """Try to merge the column transformer to a simple transformer."""
-    transformers = column_transformer.transformers_
-
-    assert len(transformers) > 0
-    _, transformer_0, column_0 = transformers[0]
-    columns = [column_0]
-    for _, transformer, column in transformers[1:]:
-        # all transformers are the same
-        if transformer != transformer_0:
-            return column_transformer
-        columns.append(column)
-    # all feature columns are transformed
-    if sorted(
-        [cast(str, feature_column.name) for feature_column in bq_model.feature_columns]
-    ) == sorted(columns):
-        return transformer_0
-
-    return column_transformer
diff --git a/bigframes/session/__init__.py b/bigframes/session/__init__.py
@@ -953,7 +953,7 @@ def read_gbq_model(self, model_name: str):
                 to load from the default project.
 
         Returns:
-            A bigframes.ml Model wrapping the model.
+            A bigframes.ml Model, Transformer or Pipeline wrapping the model.
         """
         import bigframes.ml.loader
 
diff --git a/tests/system/large/ml/test_compose.py b/tests/system/large/ml/test_compose.py
diff --git a/tests/system/large/ml/test_pipeline.py b/tests/system/large/ml/test_pipeline.py
