DanielLin1986
commented
3 years ago Hi there,
Thanks for your comments.
I used a Python file to extract C functions from a C file. The Python file can be found from the following link: https://github.com/DanielLin1986/function_representation_learning/blob/master/Code/ExtractCFunctionByName_v2.py. However, this file is buggy and it requires you to know the name of the function to be extracted. You can use a static tool to obtain all the names of functions in a C file, then use the function names to extract function bodies. Due to the file being buggy, the extracted function code may have issues (i.e., having extra "}" or missing return types ). Further processing is needed.
To obtain representations from a hidden layer of a neural network can be achieved using the following code (tested on Tensorflow 1.14 and 1.15 with Keras 2.2.4):
def ObtainRepresentations_by_batch_size(input_sequences, layer_number, model, BATCH_SIZE):
num_batches_per_epoch = int((len(input_sequences) - 1) / BATCH_SIZE) + 1
data_size = len(input_sequences)
representations_total = []
for batch_num in range(num_batches_per_epoch):
start_index = batch_num * BATCH_SIZE
end_index = min((batch_num + 1) * BATCH_SIZE, data_size)
print ("-------start_index------------")
print (start_index)
print ("-------end_index------------")
print (end_index)
layered_model = Model(inputs = model.input, outputs=model.layers[layer_number].output)
representations = layered_model.predict(input_sequences[start_index: end_index])
representations_total = representations_total + representations.tolist()
return np.asarray(representations_total)
YinHuiLin
Hi, I read your paper and think it would be very helpful for my work. I want to konw how to process the sard samples after ast serialization, because a c file may contains several functions. @DanielLin1986 Hope your reply, thank you!