chainer / chainer-chemistry

Chainer Chemistry: A Library for Deep Learning in Biology and Chemistry

MIT License

629 stars 130 forks source link

Graph Warp Module for Chainer Chemistry #321

Closed k-ishiguro closed 5 years ago

k-ishiguro commented 5 years ago

Pull Request for implementing the Graph Warp Module (https://arxiv.org/abs/1902.01020)

Main issues:

implement exmples of GWM for NFP, GGNN, RSGCN, and GIN
implement a naive model of GIN (Graph Isomorphism Network)
update train_molnet.py and predict_molnet.py in several ways
- enable to save and load a model snapshot on the "best validation score" epoch
- more stats reporting during training
- more stats reporting at evaluation (file dumped)
- a visualization script

codecov-io commented 5 years ago

Codecov Report

Merging #321 into master will decrease coverage by 9.17%. The diff coverage is 27.41%.

@@            Coverage Diff             @@
##           master     #321      +/-   ##
==========================================
- Coverage   89.59%   80.42%   -9.18%     
==========================================
  Files         182      197      +15     
  Lines        8272     8994     +722     
==========================================
- Hits         7411     7233     -178     
- Misses        861     1761     +900

corochann commented 5 years ago

Can we move construct_supernode_feature to common method if these are used among several preprocessors?
docstring: please use google format
Please use non-abbreviated import name: import numpy as np --> import numpy
Let me discuss about license for gwm.py & gin.py

k-ishiguro commented 5 years ago

OK I will work on these issues one by one.

k-ishiguro commented 5 years ago

Hi, I fixed the issues pointed out. please have a look at updates.

corochann commented 5 years ago

LGTM, thank you! We add tests, refactor names later!