added Instances.cv_splits method to generate list of train/test tuples as used by cross-validation

Author: fracpete

CHANGES.rst

@@ -6,6 +6,8 @@ Changelog
 
 - added `logging_level` parameter to the `start` method of the `weka.core.jvm` module, enabling the user
   to turn off debugging output in an easy way (https://github.com/fracpete/python-weka-wrapper3/issues/40)
+- added method `cv_splits` to class `Instances` from module `weka.core.dataset` to return a list of
+  train/test tuples as used by cross-validation
 - ...
 
 

python/weka/core/dataset.py

@@ -17,7 +17,7 @@
 import javabridge
 import logging
 import numpy as np
-from weka.core.classes import JavaObject
+from weka.core.classes import JavaObject, Random
 import weka.core.typeconv as typeconv
 
 # logging setup
@@ -558,6 +558,36 @@ def train_test_split(self, percentage, rnd=None):
         test_inst = Instances.copy_instances(data, train_size, test_size)
         return train_inst, test_inst
 
+    def cv_splits(self, folds=10, rnd=None, stratify=True):
+        """
+        Generates a list of train/test pairs used in cross-validation.
+        Creates a copy of the dataset beforehand when randomizing.
+
+        :param folds: the number of folds to use, >= 2
+        :type folds: int
+        :param rnd: the random number generator to use for randomization, skips randomization if None
+        :type rnd: Random
+        :param stratify: whether to stratify the data after randomization
+        :type stratify: bool
+        :return: the list of train/test split tuples
+        :rtype: list
+        """
+        result = []
+        if rnd is not None:
+            data = Instances.copy_instances(self)
+            data.randomize(rnd)
+            if stratify:
+                data.stratify(folds)
+        else:
+            data = self
+
+        for i in range(folds):
+            train = data.train_cv(folds, i, random=rnd)
+            test = data.test_cv(folds, i)
+            result.append((train, test))
+
+        return result
+
     @classmethod
     def summary(cls, inst):
         """