move the handling for contiguous C data into unify_columns

Author: paulbkoch

python/interpret-core/interpret/glassbox/_ebm/_bin.py

@@ -64,11 +64,6 @@ def eval_terms(X, n_samples, feature_names_in, feature_types_in, bins, term_feat
                 # TODO: we could pass out a bool array instead of objects for this function only
                 bad = bad != _none_ndarray
 
-            if not X_col.flags.c_contiguous:
-                # we requrested this feature, so at some point we're going to call discretize,
-                # which requires contiguous memory
-                X_col = X_col.copy()
-
             for requirements in all_requirements:
                 term_idx = requirements[-1]
                 feature_idxs = term_features[term_idx]

python/interpret-core/interpret/glassbox/_ebm/_ebm.py

@@ -873,7 +873,7 @@ def fit(self, X, y, sample_weight=None, bags=None, init_score=None):
                     msg = "A value in bags is outside the valid range -128 to 127"
                     _log.error(msg)
                     raise ValueError(msg)
-                bag = bag.astype(np.int8, copy=not bag.flags.c_contiguous)
+                bag = bag.astype(np.int8, "C", copy=False)
 
             rngs.append(bagged_rng)
             internal_bags.append(bag)

python/interpret-core/interpret/utils/_clean_x.py

@@ -691,7 +691,7 @@ def _process_continuous(X_col, nonmissings):
     # called under: fit or predict
 
     if issubclass(X_col.dtype.type, np.floating):
-        X_col = X_col.astype(dtype=np.float64, copy=False)
+        X_col = X_col.astype(np.float64, "C", copy=False)
         return X_col, None
     if issubclass(X_col.dtype.type, np.integer) or issubclass(
         X_col.dtype.type, np.bool_

python/interpret-core/interpret/utils/_compressed_dataset.py

@@ -66,10 +66,6 @@ def bin_native(
         else:
             # continuous feature
 
-            if not X_col.flags.c_contiguous:
-                # X_col could be a slice that has a stride.  We need contiguous for caling into C
-                X_col = X_col.copy()
-
             X_col = native.discretize(X_col, feature_bins)
             n_bins = len(feature_bins) + 3
 
@@ -115,10 +111,6 @@ def bin_native(
         else:
             # continuous feature
 
-            if not X_col.flags.c_contiguous:
-                # X_col could be a slice that has a stride.  We need contiguous for caling into C
-                X_col = X_col.copy()
-
             X_col = native.discretize(X_col, feature_bins)
             n_bins = len(feature_bins) + 3
 

python/interpret-core/interpret/utils/_preprocessor.py

@@ -302,10 +302,6 @@ def fit(self, X, y=None, sample_weight=None):
                     _log.error(msg)
                     raise ValueError(msg)
 
-                if not X_col.flags.c_contiguous:
-                    # X_col could be a slice that has a stride.  We need contiguous for caling into C
-                    X_col = X_col.copy()
-
                 if self.binning == "private":
                     if np.isnan(X_col).any():
                         msg = "missing values in X not supported for private binning"
@@ -548,10 +544,6 @@ def transform(self, X):
                 else:
                     # continuous feature
 
-                    if not X_col.flags.c_contiguous:
-                        # X_col could be a slice that has a stride.  We need contiguous for caling into C
-                        X_col = X_col.copy()
-
                     X_col = native.discretize(X_col, bins)
 
                     if bad is not None: