Thanks, the code also use the python random module, not just numpy random. Can you change the code to only use numpy random module? That could remove the not deterministic behavior you are seeing.

@nouiz thanks, it seems like you are right, mixing python number generator and numpy's seem to cause some non-deterministic behaviors. Changing python's random number generator to numpy's seem to fix the problem with this script.

The random come from the fac that they are independent rng. seeding python rng could have also fixed this. But I prefer to use only one if there isn't a good reason. I'll merge this PR when travis pass.

thanks

Hi @nouiz I was already seeding the python rng as well. That was the strange part.

Here is the line that I set the seed in my prev commit:
https://github.com/caglar/DeepLearningTutorials/blob/423f54d2daa67c02196737f17fc4365df6de76a1/code/lstm.py#L22

Anyways the tests are passing now.