Awesome

equivariant-benchmark

Benchmarking equivariant neural networks. This was the package used for https://arxiv.org/abs/2008.08461

You have to install exactly the following commits if you want to make this work.

SchNetPack commit information:
3c58fd1a0b9fa2b046a88e89eb0d0c9051973046  
* master 3c58fd1 [origin/master] removed the none agg mode  
origin	https://github.com/bkmi/schnetpack.git  
... which can be found at exactly this commit https://github.com/bkmi/schnetpack/commit/3c58fd1a0b9fa2b046a88e89eb0d0c9051973046

e3nn commit information:  
d60242f83e4bb6e9359c555ec03e7325802fe78e  
* master d60242f [origin/master] fixed so3 import error  
origin	https://github.com/bkmi/e3nn.git  
... which can be found at exactly this commit https://github.com/bkmi/e3nn/commit/d60242f83e4bb6e9359c555ec03e7325802fe78e  
(If you want the cannonical version of e3nn, try right after this pull request https://github.com/e3nn/e3nn/pull/64#)

Alternative

My collaborator made a more useable version. Give it a shot! https://github.com/mariogeiger/e3nn_little/blob/main/examples/qm9.py