Right now we use Dropout(0.5) in a number of places in the code. We should explore both where to put the dropout layers, as well as what percentage to drop out, and make a nice pretty graph about it. Here's a relevant comment thread about dropout layers:
Right now we use
Dropout(0.5)in a number of places in the code. We should explore both where to put the dropout layers, as well as what percentage to drop out, and make a nice pretty graph about it. Here's a relevant comment thread about dropout layers:https://www.reddit.com/r/MachineLearning/comments/42nnpe/why_do_i_never_see_dropout_applied_in/