Cross validation

This MR adds the following

• Adds cross-validation using a validation set; the best model based on the lowest validation loss is saved automatically now
• Adds trainer, extractor, training script and unit tests for FASNet architecture
• Small change in MCCNN architecture; added a flag for selecting whether to use sigmoid in eval phase.

bob/learn/pytorch/test/test.py

@@ -394,7 +394,7 @@ def test_extractors():
   from ..extractor.image import FASNetExtractor
   extractor = FASNetExtractor()
   # this architecture expects RGB images of size 3x224x224 channel images
-  data = numpy.random.rand(3, 224, 224).astype("float32")
+  data = numpy.random.rand(3, 224, 224).astype("uint8")
   output = extractor(data)
   assert output.shape[0] == 1