ENH Better error message when refit=True. (scikit-learn#7234)

Author: raghavrv

sklearn/model_selection/_search.py

@@ -24,6 +24,7 @@
 from ..base import MetaEstimatorMixin
 from ._split import check_cv
 from ._validation import _fit_and_score
+from ..exceptions import NotFittedError
 from ..externals.joblib import Parallel, delayed
 from ..externals import six
 from ..utils import check_random_state
@@ -414,6 +415,15 @@ def score(self, X, y=None):
  % self.best_estimator_)
  return self.scorer_(self.best_estimator_, X, y)
 
+ def _check_is_fitted(self, method_name):
+ if not self.refit:
+ raise NotFittedError(('This GridSearchCV instance was initialized '
+ 'with refit=False. %s is '
+ 'available only after refitting on the best '
+ 'parameters. ') % method_name)
+ else:
+ check_is_fitted(self, 'best_estimator_')
+
  @if_delegate_has_method(delegate='estimator')
  def predict(self, X):
  """Call predict on the estimator with the best found parameters.
@@ -428,6 +438,7 @@ def predict(self, X):
  underlying estimator.
 
  """
+ self._check_is_fitted('predict')
  return self.best_estimator_.predict(X)
 
  @if_delegate_has_method(delegate='estimator')
@@ -444,6 +455,7 @@ def predict_proba(self, X):
  underlying estimator.
 
  """
+ self._check_is_fitted('predict_proba')
  return self.best_estimator_.predict_proba(X)
 
  @if_delegate_has_method(delegate='estimator')
@@ -460,6 +472,7 @@ def predict_log_proba(self, X):
  underlying estimator.
 
  """
+ self._check_is_fitted('predict_log_proba')
  return self.best_estimator_.predict_log_proba(X)
 
  @if_delegate_has_method(delegate='estimator')
@@ -476,6 +489,7 @@ def decision_function(self, X):
  underlying estimator.
 
  """
+ self._check_is_fitted('decision_function')
  return self.best_estimator_.decision_function(X)
 
  @if_delegate_has_method(delegate='estimator')
@@ -492,11 +506,12 @@ def transform(self, X):
  underlying estimator.
 
  """
+ self._check_is_fitted('transform')
  return self.best_estimator_.transform(X)
 
  @if_delegate_has_method(delegate='estimator')
  def inverse_transform(self, Xt):
- """Call inverse_transform on the estimator with the best found parameters.
+ """Call inverse_transform on the estimator with the best found params.
 
  Only available if the underlying estimator implements
  ``inverse_transform`` and ``refit=True``.
@@ -508,6 +523,7 @@ def inverse_transform(self, Xt):
  underlying estimator.
 
  """
+ self._check_is_fitted('inverse_transform')
  return self.best_estimator_.transform(Xt)
 
  def _fit(self, X, y, labels, parameter_iterable):

sklearn/model_selection/tests/test_search.py

@@ -28,6 +28,7 @@
 
 from sklearn.externals.six.moves import zip
 from sklearn.base import BaseEstimator
+from sklearn.exceptions import NotFittedError
 from sklearn.datasets import make_classification
 from sklearn.datasets import make_blobs
 from sklearn.datasets import make_multilabel_classification
@@ -73,8 +74,10 @@ def predict(self, T):
  return T.shape[0]
 
  predict_proba = predict
+ predict_log_proba = predict
  decision_function = predict
  transform = predict
+ inverse_transform = predict
 
  def score(self, X=None, Y=None):
  if self.foo_param > 1:
@@ -268,6 +271,14 @@ def test_no_refit():
  hasattr(grid_search, "best_index_") and
  hasattr(grid_search, "best_params_"))
 
+ # Make sure the predict/transform etc fns raise meaningfull error msg
+ for fn_name in ('predict', 'predict_proba', 'predict_log_proba',
+ 'transform', 'inverse_transform'):
+ assert_raise_message(NotFittedError,
+ ('refit=False. %s is available only after '
+ 'refitting on the best parameters' % fn_name),
+ getattr(grid_search, fn_name), X)
+
 
 def test_grid_search_error():
  # Test that grid search will capture errors on data with different length