fix: stop ignoring arguments to MatrixFactorization.score(X, y) (#1726)

* fix: stop ignoring arguments to `MatrixFactorization.score(X, y)` * fix unit tests

Author: tswast

bigframes/ml/decomposition.py

@@ -360,5 +360,12 @@ def score(
  if not self._bqml_model:
  raise RuntimeError("A model must be fitted before score")
 
- # TODO(b/291973741): X param is ignored. Update BQML supports input in ML.EVALUATE.
- return self._bqml_model.evaluate()
+ if X is not None and y is not None:
+ X, y = utils.batch_convert_to_dataframe(
+ X, y, session=self._bqml_model.session
+ )
+ input_data = X.join(y, how="outer")
+ else:
+ input_data = X
+
+ return self._bqml_model.evaluate(input_data)

tests/system/large/ml/test_decomposition.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import pandas as pd
+import pandas.testing
 
 from bigframes.ml import decomposition
 from tests.system import utils
@@ -193,7 +194,16 @@ def test_decomposition_mf_configure_fit_load(
  )
  )
 
- reloaded_model.score(new_ratings)
+ # Make sure the input to score is not ignored.
+ scores_training_data = reloaded_model.score().to_pandas()
+ scores_new_ratings = reloaded_model.score(new_ratings).to_pandas()
+ pandas.testing.assert_index_equal(
+ scores_training_data.columns, scores_new_ratings.columns
+ )
+ assert (
+ scores_training_data["mean_squared_error"].iloc[0]
+ != scores_new_ratings["mean_squared_error"].iloc[0]
+ )
 
  result = reloaded_model.predict(new_ratings).to_pandas()
 

tests/unit/ml/test_golden_sql.py

@@ -81,6 +81,7 @@ def mock_X(mock_y, mock_session):
  ["index_column_id"],
  ["index_column_label"],
  )
+ type(mock_X).sql = mock.PropertyMock(return_value="input_X_sql_property")
  mock_X.reset_index(drop=True).cache().sql = "input_X_no_index_sql"
  mock_X.join(mock_y).sql = "input_X_y_sql"
  mock_X.join(mock_y).cache.return_value = mock_X.join(mock_y)
@@ -248,7 +249,7 @@ def test_decomposition_mf_predict(mock_session, bqml_model, mock_X):
  )
 
 
-def test_decomposition_mf_score(mock_session, bqml_model, mock_X):
+def test_decomposition_mf_score(mock_session, bqml_model):
  model = decomposition.MatrixFactorization(
  num_factors=34,
  feedback_type="explicit",
@@ -258,8 +259,23 @@ def test_decomposition_mf_score(mock_session, bqml_model, mock_X):
  l2_reg=9.83,
  )
  model._bqml_model = bqml_model
- model.score(mock_X)
-
+ model.score()
  mock_session.read_gbq.assert_called_once_with(
  "SELECT * FROM ML.EVALUATE(MODEL `model_project`.`model_dataset`.`model_id`)"
  )
+
+
+def test_decomposition_mf_score_with_x(mock_session, bqml_model, mock_X):
+ model = decomposition.MatrixFactorization(
+ num_factors=34,
+ feedback_type="explicit",
+ user_col="user_id",
+ item_col="item_col",
+ rating_col="rating_col",
+ l2_reg=9.83,
+ )
+ model._bqml_model = bqml_model
+ model.score(mock_X)
+ mock_session.read_gbq.assert_called_once_with(
+ "SELECT * FROM ML.EVALUATE(MODEL `model_project`.`model_dataset`.`model_id`,\n (input_X_sql_property))"
+ )

third_party/bigframes_vendored/sklearn/decomposition/_mf.py

@@ -73,11 +73,13 @@ def score(self, X=None, y=None):
  for the outputs relevant to this model type.
 
  Args:
- X (default None):
- Ignored.
+ X (bigframes.dataframe.DataFrame | bigframes.series.Series | None):
+ DataFrame of shape (n_samples, n_features). Test samples.
+
+ y (bigframes.dataframe.DataFrame | bigframes.series.Series | None):
+ DataFrame of shape (n_samples,) or (n_samples, n_outputs). True
+ labels for `X`.
 
- y (default None):
- Ignored.
  Returns:
  bigframes.dataframe.DataFrame: DataFrame that represents model metrics.
  """