sparse_attentive_backtracking_release

Code for our paper "Sparse Attentive Backtracking: Sparse Attentive Backtracking: Temporal Credit Assignment Through Reminding" https://papers.nips.cc/paper/7991-sparse-attentive-backtracking-temporal-credit-assignment-through-reminding.pdf

Create environment using

conda env create -f env.yml

Need:

• pytorch 0.1.12
• CUDA 8.0
• cuDNN 6

To run experiment, use the following command

python train_copying_torch.py -T 100 --model sparseattn --trunc 5 --topk 5