Correctness of an implementation of a Paper

Hey guys,
I have tried to implement the CNN-LSTM in the paper, could someone help me out with the correctness of the implementation.Their implementation in tensorflow is available- for anybody who can understand that. The parameters to layers have been given just to see if the model works. The test loss looks like a step function and the train and test accuracy is always 50% through any epoch, so please help me out.
Here’s a link to my code