ENH Uses binned values from training to find missing values (#16883)

thomasjpfan · web-flow · commit e5cc2b0e20df · 2020-06-25T14:24:12.000+02:00
* ENH Uses training data to find missing values

* CLN Uses the binned data to find missing value
diff --git a/sklearn/ensemble/_hist_gradient_boosting/gradient_boosting.py b/sklearn/ensemble/_hist_gradient_boosting/gradient_boosting.py
@@ -192,8 +192,6 @@ def fit(self, X, y, sample_weight=None):
             X_train, y_train, sample_weight_train = X, y, sample_weight
             X_val = y_val = sample_weight_val = None
 
-        has_missing_values = np.isnan(X_train).any(axis=0).astype(np.uint8)
-
         # Bin the data
         # For ease of use of the API, the user-facing GBDT classes accept the
         # parameter max_bins, which doesn't take into account the bin for
@@ -211,6 +209,11 @@ def fit(self, X, y, sample_weight=None):
         else:
             X_binned_val = None
 
+        # Uses binned data to check for missing values
+        has_missing_values = (
+            X_binned_train == self.bin_mapper_.missing_values_bin_idx_).any(
+                axis=0).astype(np.uint8)
+
         if self.verbose:
             print("Fitting gradient boosted rounds:")
 
