Some question about gender classification

[Jason-xin]

I'm so sorry to disturb you, there are some question about your comment in https://github.com/dpressel/rude-carnie/issues/36, I met the same result（91.8% in default modle）, can you tell me the reason?

Surprisely, I run the eval.py once again today, the result drops to 85%, nothing was changed before, and evaluation in my own test datasets stay the same result.

So sorry to disturb you.....