fix: reloaded transformer .transform error (#569)

* fix: reloaded transformer .transform error

* fix mypy

Author: GarrettWu

bigframes/ml/compose.py

@@ -115,14 +115,17 @@ 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()
 
+        output_names = []
         for transform_col in bq_model._properties["transformColumns"]:
+            transform_col_dict = cast(dict, transform_col)
             # pass the columns that are not transformed
-            if "transformSql" not in transform_col:
+            if "transformSql" not in transform_col_dict:
                 continue
-            transform_sql: str = cast(dict, transform_col)["transformSql"]
+            transform_sql: str = transform_col_dict["transformSql"]
             if not transform_sql.startswith("ML."):
                 continue
 
+            output_names.append(transform_col_dict["name"])
             found_transformer = False
             for prefix in _BQML_TRANSFROM_TYPE_MAPPING:
                 if transform_sql.startswith(prefix):
@@ -141,7 +144,10 @@ def camel_to_snake(name):
                     f"Unsupported transformer type. {constants.FEEDBACK_LINK}"
                 )
 
-        return cls(transformers=transformers)
+        transformer = cls(transformers=transformers)
+        transformer._output_names = output_names
+
+        return transformer
 
     def _merge(
         self, bq_model: bigquery.Model
@@ -164,6 +170,7 @@ def _merge(
                 for feature_column in bq_model.feature_columns
             ]
         ) == sorted(columns):
+            transformer_0._output_names = self._output_names
             return transformer_0
 
         return self

tests/system/large/ml/test_compose.py

@@ -142,3 +142,26 @@ def test_columntransformer_save_load(new_penguins_df, dataset_id):
     ]
     assert reloaded_transformer.transformers_ == expected
     assert reloaded_transformer._bqml_model is not None
+
+    result = transformer.fit_transform(
+        new_penguins_df[["species", "culmen_length_mm", "flipper_length_mm"]]
+    ).to_pandas()
+
+    expected = pandas.DataFrame(
+        {
+            "onehotencoded_species": [
+                [{"index": 1, "value": 1.0}],
+                [{"index": 1, "value": 1.0}],
+                [{"index": 2, "value": 1.0}],
+            ],
+            "standard_scaled_culmen_length_mm": [
+                1.313249,
+                -0.20198,
+                -1.111118,
+            ],
+            "standard_scaled_flipper_length_mm": [1.251098, -1.196588, -0.054338],
+        },
+        index=pandas.Index([1633, 1672, 1690], dtype="Int64", name="tag_number"),
+    )
+
+    pandas.testing.assert_frame_equal(result, expected, rtol=0.1, check_dtype=False)

tests/system/small/ml/test_preprocessing.py

@@ -58,7 +58,7 @@ def test_standard_scaler_normalizes(penguins_df_default_index, new_penguins_df):
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_standard_scaler_normalizeds_fit_transform(new_penguins_df):
@@ -82,7 +82,7 @@ def test_standard_scaler_normalizeds_fit_transform(new_penguins_df):
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_standard_scaler_series_normalizes(penguins_df_default_index, new_penguins_df):
@@ -110,7 +110,7 @@ def test_standard_scaler_series_normalizes(penguins_df_default_index, new_pengui
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_standard_scaler_save_load(new_penguins_df, dataset_id):
@@ -125,6 +125,22 @@ def test_standard_scaler_save_load(new_penguins_df, dataset_id):
     assert isinstance(reloaded_transformer, preprocessing.StandardScaler)
     assert reloaded_transformer._bqml_model is not None
 
+    result = reloaded_transformer.transform(
+        new_penguins_df[["culmen_length_mm", "culmen_depth_mm", "flipper_length_mm"]]
+    ).to_pandas()
+
+    expected = pd.DataFrame(
+        {
+            "standard_scaled_culmen_length_mm": [1.313249, -0.20198, -1.111118],
+            "standard_scaled_culmen_depth_mm": [1.17072, -1.272416, 0.101848],
+            "standard_scaled_flipper_length_mm": [1.251089, -1.196588, -0.054338],
+        },
+        dtype="Float64",
+        index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
+    )
+
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
+
 
 def test_max_abs_scaler_normalizes(penguins_df_default_index, new_penguins_df):
     # TODO(http://b/292431644): add a second test that compares output to sklearn.preprocessing.MaxAbsScaler, when BQML's change is in prod.
@@ -157,7 +173,7 @@ def test_max_abs_scaler_normalizes(penguins_df_default_index, new_penguins_df):
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_max_abs_scaler_normalizeds_fit_transform(new_penguins_df):
@@ -176,7 +192,7 @@ def test_max_abs_scaler_normalizeds_fit_transform(new_penguins_df):
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_max_abs_scaler_series_normalizes(penguins_df_default_index, new_penguins_df):
@@ -199,7 +215,7 @@ def test_max_abs_scaler_series_normalizes(penguins_df_default_index, new_penguin
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_max_abs_scaler_save_load(new_penguins_df, dataset_id):
@@ -214,6 +230,22 @@ def test_max_abs_scaler_save_load(new_penguins_df, dataset_id):
     assert isinstance(reloaded_transformer, preprocessing.MaxAbsScaler)
     assert reloaded_transformer._bqml_model is not None
 
+    result = reloaded_transformer.transform(
+        new_penguins_df[["culmen_length_mm", "culmen_depth_mm", "flipper_length_mm"]]
+    ).to_pandas()
+
+    expected = pd.DataFrame(
+        {
+            "max_abs_scaled_culmen_length_mm": [1.0, 0.974684, 0.959494],
+            "max_abs_scaled_culmen_depth_mm": [1.0, 0.914894, 0.962766],
+            "max_abs_scaled_flipper_length_mm": [1.0, 0.923469, 0.959184],
+        },
+        dtype="Float64",
+        index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
+    )
+
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
+
 
 def test_min_max_scaler_normalized_fit_transform(new_penguins_df):
     scaler = preprocessing.MinMaxScaler()
@@ -231,7 +263,7 @@ def test_min_max_scaler_normalized_fit_transform(new_penguins_df):
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_min_max_scaler_series_normalizes(penguins_df_default_index, new_penguins_df):
@@ -255,7 +287,7 @@ def test_min_max_scaler_series_normalizes(penguins_df_default_index, new_penguin
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_min_max_scaler_normalizes(penguins_df_default_index, new_penguins_df):
@@ -290,7 +322,7 @@ def test_min_max_scaler_normalizes(penguins_df_default_index, new_penguins_df):
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_min_max_scaler_save_load(new_penguins_df, dataset_id):
@@ -305,6 +337,22 @@ def test_min_max_scaler_save_load(new_penguins_df, dataset_id):
     assert isinstance(reloaded_transformer, preprocessing.MinMaxScaler)
     assert reloaded_transformer._bqml_model is not None
 
+    result = reloaded_transformer.fit_transform(
+        new_penguins_df[["culmen_length_mm", "culmen_depth_mm", "flipper_length_mm"]]
+    ).to_pandas()
+
+    expected = pd.DataFrame(
+        {
+            "min_max_scaled_culmen_length_mm": [1.0, 0.375, 0.0],
+            "min_max_scaled_culmen_depth_mm": [1.0, 0.0, 0.5625],
+            "min_max_scaled_flipper_length_mm": [1.0, 0.0, 0.466667],
+        },
+        dtype="Float64",
+        index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
+    )
+
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
+
 
 def test_k_bins_discretizer_normalized_fit_transform_default_params(new_penguins_df):
     discretizer = preprocessing.KBinsDiscretizer(strategy="uniform")
@@ -322,7 +370,7 @@ def test_k_bins_discretizer_normalized_fit_transform_default_params(new_penguins
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_k_bins_discretizer_series_normalizes(
@@ -344,7 +392,7 @@ def test_k_bins_discretizer_series_normalizes(
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_k_bins_discretizer_normalizes(penguins_df_default_index, new_penguins_df):
@@ -374,7 +422,7 @@ def test_k_bins_discretizer_normalizes(penguins_df_default_index, new_penguins_d
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_k_bins_discretizer_normalizes_different_params(
@@ -406,7 +454,7 @@ def test_k_bins_discretizer_normalizes_different_params(
         index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
     )
 
-    pd.testing.assert_frame_equal(result, expected, rtol=1e-3)
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
 
 
 def test_k_bins_discretizer_save_load(new_penguins_df, dataset_id):
@@ -423,6 +471,22 @@ def test_k_bins_discretizer_save_load(new_penguins_df, dataset_id):
     assert reloaded_transformer.strategy == transformer.strategy
     assert reloaded_transformer._bqml_model is not None
 
+    result = reloaded_transformer.fit_transform(
+        new_penguins_df[["culmen_length_mm", "culmen_depth_mm", "flipper_length_mm"]]
+    ).to_pandas()
+
+    expected = pd.DataFrame(
+        {
+            "kbinsdiscretizer_culmen_length_mm": ["bin_6", "bin_4", "bin_2"],
+            "kbinsdiscretizer_culmen_depth_mm": ["bin_6", "bin_2", "bin_5"],
+            "kbinsdiscretizer_flipper_length_mm": ["bin_6", "bin_2", "bin_4"],
+        },
+        dtype="string[pyarrow]",
+        index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
+    )
+
+    pd.testing.assert_frame_equal(result, expected, rtol=0.1)
+
 
 def test_one_hot_encoder_default_params(new_penguins_df):
     encoder = preprocessing.OneHotEncoder()
@@ -560,6 +624,29 @@ def test_one_hot_encoder_save_load(new_penguins_df, dataset_id):
     assert reloaded_transformer.max_categories == transformer.max_categories
     assert reloaded_transformer._bqml_model is not None
 
+    result = reloaded_transformer.fit_transform(
+        new_penguins_df[["species", "sex"]]
+    ).to_pandas()
+
+    expected = pd.DataFrame(
+        {
+            "onehotencoded_species": [
+                [{"index": 1, "value": 1.0}],
+                [{"index": 1, "value": 1.0}],
+                [{"index": 2, "value": 1.0}],
+            ],
+            "onehotencoded_sex": [
+                [{"index": 2, "value": 1.0}],
+                [{"index": 1, "value": 1.0}],
+                [{"index": 1, "value": 1.0}],
+            ],
+        },
+        dtype=ONE_HOT_ENCODED_DTYPE,
+        index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
+    )
+
+    pd.testing.assert_frame_equal(result, expected)
+
 
 def test_label_encoder_default_params(new_penguins_df):
     encoder = preprocessing.LabelEncoder()
@@ -677,5 +764,21 @@ def test_label_encoder_save_load(new_penguins_df, dataset_id):
     assert reloaded_transformer.max_categories == transformer.max_categories
     assert reloaded_transformer._bqml_model is not None
 
+    result = reloaded_transformer.transform(new_penguins_df).to_pandas()
+
+    expected = pd.DataFrame(
+        {
+            "labelencoded_species": [
+                1,
+                1,
+                2,
+            ],
+        },
+        dtype="Int64",
+        index=pd.Index([1633, 1672, 1690], name="tag_number", dtype="Int64"),
+    )
+
+    pd.testing.assert_frame_equal(result, expected)
+
 
 # TODO(garrettwu): add OneHotEncoder tests to compare with sklearn.