1. formatting data files:
cd ./data/dnn_ordered_traces/
make
# now only formatting is deleting the first column
# you can check ./data/dnn_ordered_traces/*.num to confirm
# maps for instructions and data are stored in ./data/dnn_ordered_traces/*.num.instr_map and ./data/dnn_ordered_traces/*.num.data_map, respectively.
# But the maps were already taken care of. ( for now only instr_map is used )
2. train: (e.g. dataset is gcc)
make gcc.train
# model files will be stored in ./gcc/
3. test with test file:
make gcc.test
# prediction will be stored in ./gcc.pred
4. test each sample:
check main.py, a demo is written in lines 122-190:
pseudo code:
0) read LSTM model by saver.restore(...)
1) read batches of samples from test_reader (created from test file "./data/dnn_ordered_traces/gcc.num.test", then translated by instruction map )
2) at each step, feed one sample into LSTM model by model.predict(...)
3) do initialization by model.init()
You can modify this part of code.