reorganization, make into interfaces...

[rkurchin]

This is currently a WIP where I'm implementing a bunch of the stuff we've discussed vis a vis splitting out functionalities, "interface-ifying" things, and further separating concerns wrt the codec syntax. Some broad remarks..

• design philosophy is to be much more lightweight than before and allow user more flexibility on what is loaded. CF becomes primarily interfaces and a lot of concrete functionality is moved to other packages (e.g. featurizations to specific models e.g. https://github.com/Chemellia/AtomicGraphNets.jl/pull/96, graph-building to AtomGraphs)
• perspective on submodules – IMO they’re primarily to avoid namespace cluttering/conflicts. Hence, removed a lot that don’t create conflicts and export things that I want always to be available (codecs, abstract types, etc.) and added some for things that some users might not need but I don’t think need to be split out into another package (EFD stuff)

OIther specific changes:

• As we discussed a few weeks ago, FD's now don't have a codec attached (so that type parameter is also removed), but the codec is an argument to the encode function as in encode(fd, atoms, codec) but the final argument is optional because it will default to default_codec(fd) if not provided
• Trying to reorganize exports to mostly be at the top level with a few exceptions of more "internal" functions

Interfaces and associated functions:

• codecs: encode, decode, output_shape
• feature descriptors: get_value, encodable_elements, default_codec
• atoms: no constraint on type so that both graphs and AtomsBase systems will work, but they need to support the elements function
• featurizations: similar functionality to before

Remaining todos:

• [ ] update tests (this is a proxy for probably finding other things I still need to fix)
• [ ] update docs

[rkurchin]

NB's for people:

• @thazhemadam this removes your OFM stuff, so that should probably get put in a separate package (since OFM is used in other models too, so I don't think it needs to be combined with the TB nets stuff)
• @DhairyaLGandhi similarly for Weave, that should get put in the WeaveModel package (and probably we should make a new package for MolecularGraph features since those could certainly see broader use than Weave

The only concrete FD I'm keeping in CF for now is the EFD since that's quite generic and also depends very little on the nature of the structure representation. But I may split out modules for different sources of lookup tables e.g. pymatgen, matminer, etc...

[thazhemadam]

@rkurchin test failures seem to be primarily because we are still using Conda in deps/build.jl, when Conda.jl was removed as a dependency in 352044ffd4ad2829c82bfc348851c56c38ded5c8.

[rkurchin]

Thanks, yeah I haven't finished fixing all the tests yet. Will keep that in mind, hoping to get to it later today!

[codecov[bot]]

Codecov Report

Merging #132 (899fbac) into main (ef30f5f) will increase coverage by 4.38%. The diff coverage is 86.72%.

:exclamation: Current head 899fbac differs from pull request most recent head 08cca74. Consider uploading reports for the commit 08cca74 to get more accurate results

@@            Coverage Diff             @@
##             main     #132      +/-   ##
==========================================
+ Coverage   80.80%   85.18%   +4.38%     
==========================================
  Files          20       11       -9     
  Lines         573      162     -411     
==========================================
- Hits          463      138     -325     
+ Misses        110       24      -86     

Impacted Files	Coverage Δ
src/ChemistryFeaturization.jl	0.00% <0.00%> (-80.00%)	:arrow_down:
src/featurizedatoms.jl	44.44% <66.66%> (+4.44%)	:arrow_up:
src/codecs/onehotonecold_utils.jl	78.94% <78.94%> (ø)
src/codecs/onehotonecold.jl	84.44% <83.33%> (+0.23%)	:arrow_up:
src/features/features.jl	80.00% <87.50%> (-8.89%)	:arrow_down:
src/codecs/codecs.jl	100.00% <100.00%> (ø)
src/codecs/directcodec.jl	100.00% <100.00%> (ø)
src/codecs/simplecodec.jl	100.00% <100.00%> (ø)
src/features/elementfeature.jl	100.00% <100.00%> (+15.00%)	:arrow_up:
src/features/elementfeature_utils.jl	100.00% <100.00%> (ø)
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update ef30f5f...08cca74. Read the comment docs.

[rkurchin]

I'd like to go ahead and get this merged so that I can register and then merge/register the associated AtomGraphs and AGN updates (and @DhairyaLGandhi can do so with WeaveModel as well). Once everything is registered I'll be able to fix the docs issues so that they can build again.

Anyway, would love any additional feedback you guys have here! I think we were generally in agreement about these changes. so I may just merge in a day or so if I don't hear any objections.

[rkurchin]

I'm generally on board with the utils/ directory idea, but it doesn't seem necessary to me when there would only be one file in it. I think if/when we add more such things we can introduce that level to the hierarchy.