Fix xgboost feature selector callbacks ...

src/slickml/selection/_xgboost.py

@@ -1045,6 +1045,10 @@ def _callbacks(self) -> None:
  None
  """
  if self.callbacks:
+ # TODO(amir): ditch print with logger
+ print(
+ "Warning: The `cv` will break if the `early_stopping_rounds` criterion was not satisfied."
+ )
  self.callbacks = [
  xgb.callback.EvaluationMonitor(
  rank=0,

tests/slickml/selection/test_xgboost.py

@@ -237,7 +237,6 @@ def test_reg_xgboostfeatureselector__passes__with_defaults(
  assert_that(cv_results_fig).is_instance_of(Figure)
  assert_that(feature_frequency_fig).is_instance_of(Figure)
 
- # TODO(amir): change the code for `callback=True` and add the unit-test
  @pytest.mark.parametrize(
  ("clf_x_y"),
  [
@@ -257,14 +256,13 @@ def test_clf_xgboostfeatureselector__passes__with_valid_inputs(
  ) -> None:
  """Validates `XGBoostFeatureSelector` instanation passes with valid inputs for classification."""
  X, y = clf_x_y
- # TODO(amir): callbacks=True has not been tested yet. The look of `self._cv()` would change
- # therefore the logic for `if _feature_gain["feature"].str.contains("noisy").sum() != 0:`
- # should be changed accordingly
  xfs = XGBoostFeatureSelector(
  n_iter=1,
  sparse_matrix=True,
  scale_std=True,
  params={"eval_metric": "logloss"},
+ early_stopping_rounds=1000,
+ callbacks=True,
  )
  xfs.fit(X, y)
  params = xfs.get_params()
@@ -291,7 +289,7 @@ def test_clf_xgboostfeatureselector__passes__with_valid_inputs(
  assert_that(xfs.nth_noise_threshold).is_instance_of(int)
  assert_that(xfs.nth_noise_threshold).is_equal_to(1)
  assert_that(xfs.early_stopping_rounds).is_instance_of(int)
- assert_that(xfs.early_stopping_rounds).is_equal_to(20)
+ assert_that(xfs.early_stopping_rounds).is_equal_to(1000)
  assert_that(xfs.random_state).is_instance_of(int)
  assert_that(xfs.random_state).is_equal_to(1367)
  assert_that(xfs.metrics).is_instance_of(str)
@@ -304,7 +302,7 @@ def test_clf_xgboostfeatureselector__passes__with_valid_inputs(
  assert_that(xfs.shuffle).is_true()
  assert_that(xfs.verbose_eval).is_instance_of(bool)
  assert_that(xfs.verbose_eval).is_false()
- assert_that(xfs.callbacks).is_none()
+ assert_that(xfs.callbacks).is_not_none()
  assert_that(xfs.scale_mean).is_instance_of(bool)
  assert_that(xfs.scale_mean).is_false()
  assert_that(xfs.scale_std).is_instance_of(bool)
@@ -370,6 +368,8 @@ def test_reg_xgboostfeatureselector__passes__with_valid_inputs(
  sparse_matrix=True,
  scale_std=True,
  params={"eval_metric": "mae"},
+ callbacks=True,
+ early_stopping_rounds=1000,
  )
  xfs.fit(X, y)
  params = xfs.get_params()
@@ -396,7 +396,7 @@ def test_reg_xgboostfeatureselector__passes__with_valid_inputs(
  assert_that(xfs.nth_noise_threshold).is_instance_of(int)
  assert_that(xfs.nth_noise_threshold).is_equal_to(1)
  assert_that(xfs.early_stopping_rounds).is_instance_of(int)
- assert_that(xfs.early_stopping_rounds).is_equal_to(20)
+ assert_that(xfs.early_stopping_rounds).is_equal_to(1000)
  assert_that(xfs.random_state).is_instance_of(int)
  assert_that(xfs.random_state).is_equal_to(1367)
  assert_that(xfs.metrics).is_instance_of(str)
@@ -407,7 +407,7 @@ def test_reg_xgboostfeatureselector__passes__with_valid_inputs(
  assert_that(xfs.shuffle).is_true()
  assert_that(xfs.verbose_eval).is_instance_of(bool)
  assert_that(xfs.verbose_eval).is_false()
- assert_that(xfs.callbacks).is_none()
+ assert_that(xfs.callbacks).is_not_none()
  assert_that(xfs.scale_mean).is_instance_of(bool)
  assert_that(xfs.scale_mean).is_false()
  assert_that(xfs.scale_std).is_instance_of(bool)