extract features from NN, use as input to RF

[x] figure out how to extract features from a NN layer
[x] figure out how to load a NN from the weights and JSON file instead of training
[x] grab NN from Josh's GPU1 - i.e. good convnet
[x] develop RF model that takes NN inner features as input and outputs maliciousness rating, optimizing on AUC
[x] compare NN AUC to NN-RF AUC
[ ] permute e.g. layers, RF parameters to see if you can get better results
[ ] write up results

hillarysanders / deep_learning