I’m following a DL reading seminar with some PhD students. We follow the book Dive into Deep Learning but the code is written in Python. As a good exercise, and since I’m a julia user, I’m moving all the implementations to Julia. We are at the RNN chapter and I just can’t get a simple RNN to work on the basic time machine dataset The run-time is awfully slow. My model has < 30000 parameters. I think I may be miss-using the Flux API.
I am looking for text prediction implementations of RNN in Flux or a full tutorial. I only found the model zoo with the char-rnn but it’s only iterating once over the data (I don’t understand when I must use
reset!) and the dataloader is not working properly. Moving this model and data to CUDA also gives scalar indexing warnings.
Any good material you can recommend ?