Acknowledgment

This repo is based on https://github.com/Zhenye-Na/DA-RNN.

Improvements

• DataSet DataLoader implementation; easier training loop
• Batch training
• GPU support; Multi-GPUs support currently not working due to pytorch 0.4's bug
• Corrections of differences between original code and paper

Multi-GPUs BUG [Solved]

Solved in pytorch 1.0. Hooray for users!

Pytorch 0.4 Bug relates to this issue: pytorch/pytorch#7092

RuntimeError: torch/csrc/autograd/variable.cpp:115: get_grad_fn: Assertion output_nr == 0 failed.