TODO list

[YutackPark]

TODO list

----------- on 'refactor' branch && v0.10.0 --------------

• [x] Hide type_map
• [x] Hide toggle grad things
• [x] Refactoring or from scratch: scripts/processing_dataset.py and train/dataset.py
• [x] Implement universal keyword for chemical_species key
• [x] Remove from pair_e3gnn.cpp
• [x] Refactoring sevenn_inference (I bet writing from scratch is better)
• [x] #99, add model argument for calculator
• [x] #92
• [x] Remove deployed models in pretrained_potentials
• [x] Merge above changes(v0.10.0) (don't forget #73)

------------ Urgent -----------------

• [ ] Notebook tutorial
• [x] Modern test code + github actions
• [ ] documentation
• [ ] #71

----------- Future -------------

• [x] on-the-fly graph build training routine # experimentally implemented in ase_db branch
• [ ] Accelerated training with graph padding and torch.compile
• [ ] Modal select inputs for LAMMPS
• [ ] tag-wise weights (from sturcture_list or more) and logging
• [ ] Remove redundant keys in dataload.py (node_features, pos, cell, cell_shift, num_atoms(?), per_atom_energy)

[thangckt]

I suggest refactor the code to process dataset. It is too complicate, hard to read, making somebody lazy to contribute

[thangckt]

hi @YutackPark Are you consider to pure pyproject.toml ? I see many other project do not need setup.* files. I also recommand to use ruff as both linter & formatter, it demonstrates better performance over other ones.

[YutackPark]

Hi @thangckt

Are you consider to pure pyproject.toml ? I see many other project do not need setup.* files. Absolutely yes. But I left setup.py for editable mode when I first wrote project.toml. Because I wasn't able to install SevenNet with pip -e.

I also recommand to use ruff as both linter & formatter, it demonstrates better performance over other ones. Thanks for the recommendation. Maybe I could migrate to ruff with #98, and before github actions.