Crystal Support

[nissy-dev]

DONE

• implement MEGNet Model
• support Material Project Dataset
• refactoring converter method (each model has each converter method)
• migrate CGCNN Model
• confirm test
  • pass examples/test_example.sh
  • pass pytest -m "not (gpu or slow)" tests

The detail description is here

TODO

• add test (WIP)
• fix path for Material Project Dataset

[corochann]

TODO later

[codecov-io]

Codecov Report

Merging #396 into experimental_mp will decrease coverage by 7.03%.
The diff coverage is 57.57%.

@@                 Coverage Diff                 @@
##           experimental_mp     #396      +/-   ##
===================================================
- Coverage             91.3%   84.27%   -7.04%     
===================================================
  Files                  227      250      +23     
  Lines                10999    11951     +952     
===================================================
+ Hits                 10043    10072      +29     
- Misses                 956     1879     +923

[nissy-dev]

I wrote tests for CGCNN & MEGNet. Currently, all tests pass the forward test,
but don't pass the backward test except test_megnet_readout.py.

Now, most of the backward test are commented out and
I investigate why the backward test fails.

[corochann]

LGTM, let me check running the code.

[corochann]

could you change __init__ method get data_dir path in argument and load json & create feat_dict inside __init__ ?
Then we can re-use self.feat_dict in this method.

[corochann]

LGTM.
This is experimental branch, so let me merge this PR at first.
I will proceed following:

• include changes of current master branch (dataset class are added to support converter)
• write download script of materials project dataset
  • modify data_dir path etc.
• include citation/license properly
• check code runs
• merge to master branch