float to int in model creation

[pbrusco-asapp]

Hi!

Thank you very much for the project!

I've found a problem when running with the lastest versions of torch (since torch==0.4.0 is not supported by my cuda config)

In /speech_emotion_recognition/models.py
nn.Linear(self.flat_size, cfg.num_classes) causes an exception since self.flat_size is a float.

I changed the line:
self.flat_size = cfg.num_filters[len(cfg.num_filters)-1] * self.out_size**2

for
self.flat_size = int(cfg.num_filters[len(cfg.num_filters)-1] * self.out_size**2)

and it worked!

Thanks again.