LSTM/GRU-style RNN implemented in PyTorch with PyTorch Lightning for flexibility in logging and analysis. This repo adds some new datasets that are helpful for experiments and research with RNNs.
For variable-length sequences and some parameter K, this dataset is helpful for training an RNN to memorize the location to echo. It is a very simple dataset that is helpful for debugging.
0 1 2 3 4 5 -> 1
9 2 3 9 -> 2
0 7 8 2 2 0 5 3 2 -> 7
The binary addition task adds 7-bit binary numbers in order to teach the RNN to add and carry the appropriate numbers.
11001
00101
-----
11110
The Shakespeare text file dataset is also included.