batch_norm is_training parameter

[autasi]

Shouldn't the 'is_training' parameter be passed to batch_norm normalizer via 'normalizer_params' of conv2d? The same applies to the other functions in this file.

https://github.com/nnUyi/PeleeNet/blob/60d179101adebd0c5ec9853c2aa09694b3a51b79/layers.py#L19

[chyueG]

I think the writer forget this parameter,also ,in layer.py file line 79,function def _classification_layer,the writer forgot to change the dropout value when test