DL-LA_CNN_SA_inputs.py

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Reshape, Conv1D, MaxPooling1D, Flatten, Dense
from tensorflow.keras.activations import relu
from traceloader import TraceConfig
import numpy
import logging


# Specify the raw data set to be analyzed (e.g.)
traceset = 'FPGA_PRESENT_RANDOMIZED_CLOCK'
traceconfig = TraceConfig()
tracelength = traceconfig.getnrpoints(traceset)
peakdist = traceconfig.getpeakdistance(traceset)


# Define the training and validation parameters
nrtrain = 5000
nrval = 10000
nrepochs = 50
batchsize = 2000
filter = 12
kernel_mult = 2
strides = 2
pool = 2
if nrtrain > 100000:
    nrsensi = 100000
else:
    nrsensi = nrtrain
balance = 1


# Call the trace preparation function to create the required training and validation set
train_x, train_y, val_x, val_y = traceconfig.prep_traces(traceset, nrtrain, nrval, balance)


# Define the model, train the model and validate the model using the created data sets
model = Sequential(
    [Reshape((tracelength,1), input_shape = (tracelength,)),
    Conv1D(filters=filter, kernel_size=kernel_mult*peakdist, strides=peakdist//strides, input_shape=(tracelength,1), activation='relu'),
    MaxPooling1D(pool_size=pool),
    Flatten(),
    Dense(2, activation='sigmoid')])
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
out=model.fit(train_x, train_y, validation_data=(val_x, val_y), epochs=nrepochs, batch_size=batchsize, verbose=1)


# Dump the validation accuracy into an ASCII file
logging.basicConfig(filename='val_acc.log', format='%(message)s', level=logging.INFO)
logging.info("val_accuracy: {}".format(out.history['val_accuracy']))


# Perform sensitivity analysis based on the training set
inp = tf.Variable(train_x[:nrsensi], dtype=tf.float32)
with tf.GradientTape() as tape:
    tape.watch(inp)
    preds = model(inp)
    trues = tf.Variable(train_y[:nrsensi], dtype=tf.float32)
    loss = tf.keras.losses.binary_crossentropy(trues, preds)
    grads = tape.gradient(loss, inp)
grads_sum = numpy.sum(numpy.abs(grads), axis=0)


# Dump the sensitivity analysis results into a binary file
f = open("sensi.dat","wb")
f.write(grads_sum.astype("double"))
f.close()