when it determines that there have been a lot of steps without progress and
shrinks the learning rate.  Instead, it removes only the oldest 100.  The
problem with the old way of removing all the loss values in the history was
that if you set the steps without progress threshold to a really high number
you would often observe that the last few learning rate values were obviously
not making progress, however, since all the previous loss values were forgotten
the trainer needed to fully populate it's loss history from scratch before it
would figure this out.  This new style makes the trainer not waste time running
this excessive optimization of obviously useless mini-batches.

which it was in all ways except you couldn't deserialize directly like you
would expect.  This has now been fixed.

across scales regardless of the input image size.  Previously, if you gave
really large images or really small images it had a bias towards giving only
large patches or small patches respectively.

concat layer's backward() method.  It was assigning the gradient to previous
layers instead of adding the gradient, as required by the layer interface
specification.  This change also noticeably speeds up concat layers since only
one CUDA kernel launch now happens per concat operation, rather than one
kernel launch for each sample in a tensor.

switch for use in plugin environments.

renamed the objects and generally improved documentation.

Also added an operator<< for dnn_trainer that prints the parameters it's using.

These changes also break backwards compatibility with the previous
serialization format for dnn_trainer objects.