Remove for loop from datasets.supervised.SupervisedDataSet.splitWithP…

…roportion

Now splitWithProporion uses numpy array indicies with numpy.random.permutation instead of for loop, before this change on large datasets this method was very slow, now its finish almost instant.

pybrain/datasets/supervised.py

@@ -2,6 +2,7 @@
 
 __author__ = 'Thomas Rueckstiess, ruecksti@in.tum.de'
 
+from numpy import random
 from random import sample
 from scipy import isscalar
 
@@ -104,16 +105,15 @@ def evaluateModuleMSE(self, module, averageOver = 1, **args):
     def splitWithProportion(self, proportion = 0.5):
         """Produce two new datasets, the first one containing the fraction given
         by `proportion` of the samples."""
-        leftIndices = set(sample(list(range(len(self))), int(len(self)*proportion)))
-        leftDs = self.copy()
-        leftDs.clear()
-        rightDs = leftDs.copy()
-        index = 0
-        for sp in self:
-            if index in leftIndices:
-                leftDs.addSample(*sp)
-            else:
-                rightDs.addSample(*sp)
-            index += 1
+        indicies = random.permutation(len(self))
+        separator = int(len(self) * proportion)
+
+        leftIndicies = indicies[:separator]
+        rightIndicies = indicies[separator:]
+
+        leftDs = SupervisedDataSet(inp=self['input'][leftIndicies].copy(),
+                                   target=self['target'][leftIndicies].copy())
+        rightDs = SupervisedDataSet(inp=self['input'][rightIndicies].copy(),
+                                    target=self['target'][rightIndicies].copy())
         return leftDs, rightDs