training_label_type instead of a single label_type.  This way, the label
type used for training can be distinct from the type output by the network.
This change breaks backwards compatibility with the previous API.

this frustrates Visual Studio users who want to compile dlib as a static
library.

versions were calling into cuDNN, however, the cuDNN functions for doing this
are horrifically slow, well over 100x slower than they should be, which is
surprising since these functions are so trivial.

little.

NaN is encountered.

in some cases.

Added Travis badge  ![badge](https://travis-ci.org/davisking/dlib.svg?branch=master)

If temp is nonzero but a subnormal then temp2 matrix may contain infinities which may cause temp assignment in line 225 to be a NaN (as a difference of two positive infinities).