lstm for opseq

lstm_opseq.py

#do a lot of imports
from keras.models import Sequential
from keras.layers import LSTM, Dense
import numpy as np
from os import listdir
import os.path
import json
from sklearn.preprocessing import OneHotEncoder


#names of assembly instructions families
labels = ["cdt", "udt", "sdt", "adt", "cmpdt", "cvt", "bai", "iai", 
          "dai", "fai", "fci", "sai", "li", "sri", "bii", "byi",
          "cj", "uj", "int", "si", "io", "flg", "seg", "misc", "sr", 
          "rng", "arr", "pmi", "pci", "mmxt", "mmxc", "mmxa", 
          "mmxcmp", "mmxl", "mmxsr", "mmxsm", "sset", "ssea",
          "ssecmp", "ssel", "ssesu", "ssecvt", "fdt", "ftrdt", "flc",
          "tse", "ssebi", "vmx", "other"]

#encode labels into one hot vector
labels_array = np.array(labels).reshape(-1, 1)
hot_encoder = OneHotEncoder(sparse = False)
encoded_labels = hot_encoder.fit_transform(labels_array)

encode_dict = {}
for l,e in zip(labels,encoded_labels):
    encode_dict[l] = e

#configs for training
data_dim = 49  #dimension of a vector
num_classes = 2 #2 classes 
batch_size = 256 
lstm_units = 128 #number of units by layer
lstm_layers = 3 
stateful = False 
epochs = 100


#LSTM Builder 
def create_lstm():
    model = Sequential()
    layer1 = LSTM(lstm_units, return_sequences = True, input_shape = (timesteps, data_dim))
    model.add(layer1)
    for _ in range(lstm_layers -2):
        model.add(LSTM(lstm_units, return_sequences = True))
    model.add(LSTM(lstm_units))
    model.add(Dense(1, activation = 'sigmoid'))
    return model
  
def model_compiler(model):
  #you can definitely change those parameters
    loss = 'binary_crossentropy'
    optimizer = 'rmsprop'
    metrics = ['binary_accuracy']
    model.compile(loss = loss, optimizer = optimizer, metrics = metrics)
    model.summary()
    
def model_trainer(model, data, labels):
    history = model.fit(data, labels, validation_split = 0.25, epochs = 50, batch_size = 1536)
    return history, model
    return model