search

QData / spacetimeformer

Multivariate Time Series Forecasting with efficient Transformers. Code for the paper "Long-Range Transformers for Dynamic Spatiotemporal Forecasting."
https://arxiv.org/abs/2109.12218
MIT License
808 stars 191 forks source link

Request to Submit PR for Issues #69

Closed dancassin closed 1 year ago

dancassin commented 1 year ago

Followed the instructions provided on the README, and I have some potential changes if agreed upon:

  1. source activate spacetimeformer should be: conda activate spacetimeformer

  2. in order to run the training commands, it requires a second cd spacetimeformer after lines to install requirements.txt in order to run the train.py

  3. Additon of torchvision and opt_einsum to requirements.txt

jakegrigsby commented 1 year ago

I appreciate the notes, I put some of these changes into the PR that updated the readme for v3 of the arXiv paper.

Pytorch/torchvision are such key dependencies that the package assumes researchers already have them installed and does not try to install them for you. I do this to avoid dealing with CUDA version control errors that are gpu-specific.