just noticed sklearn.cross_validation.cross_val_predict

formasaurus/fieldtype_model.py

@@ -31,14 +31,14 @@
 import numpy as np
 from sklearn.grid_search import RandomizedSearchCV
 from sklearn.metrics import make_scorer
+from sklearn.cross_validation import cross_val_predict
 from sklearn_crfsuite import CRF
 from sklearn_crfsuite.metrics import flat_f1_score
 
 from formasaurus import formtype_model
 from formasaurus.html import get_fields_to_annotate, get_text_around_elems
 from formasaurus.text import (normalize, tokenize, ngrams, number_pattern,
     token_ngrams)
-from formasaurus.utils import select_by_index
 from formasaurus.annotation import get_annotation_folds
 
 
@@ -152,15 +152,8 @@ def get_realistic_form_labels(annotations, n_folds=10, model=None,
     else:
         y = np.asarray([a.type for a in annotations])
 
-    y_pred = np.empty(len(annotations), dtype=object)
-
-    for idx_train, idx_test in get_annotation_folds(annotations, n_folds):
-        X_train = select_by_index(X, idx_train)
-        X_test = select_by_index(X, idx_test)
-        model.fit(X_train, y[idx_train])
-        y_pred[idx_test] = model.predict(X_test)
-
-    return y_pred
+    folds = get_annotation_folds(annotations, n_folds)
+    return cross_val_predict(model, X, y, cv=folds)
 
 
 def get_form_features(form, form_type, field_elems=None):

formasaurus/utils.py

@@ -60,17 +60,6 @@ def inverse_mapping(dct):
     return {v:k for k,v in dct.items()}
 
 
-def select_by_index(arr, index):
-    """
-    Like numpy indexing, but for lists. This is for cases
-    conversion to numpy array is problematic.
-
-    >>> select_by_index(['a', 'b', 'c', 'd'], [0, 3])
-    ['a', 'd']
-    """
-    return [arr[i] for i in index]
-
-
 def at_root(*args):
     """ Return path relative to formasaurus source code """
     return os.path.join(os.path.dirname(__file__), *args)