[filipetessaro]

Hi, this adds an optional seed argument to the three get_train_val_test_loader functions in chgnet/data/dataset.py (GraphData, StructureJsonData and the module-level one), as suggested in #253.

With seed=None, the default, nothing changes: the split still uses the global random state, so code that calls random.seed() first gets the same split as before. With an integer, the split comes from random.Random(seed), so a retrain can rebuild the same train/val/test mp-ids without touching global state. Only the split is seeded; DataLoader shuffling inside each set is unchanged.

Added test_data_loader_seed in tests/test_dataset.py (same seed gives the same split, different seeds don't).

This doesn't recover the 0.2.0/0.3.0 test ids from the issue, of course. That part still needs the original lists if you have them.

Report: https://github.com/paper-rerun/paper-rerun/blob/main/reports/2023-deng-chgnet/REPORT.md

This change was prepared with AI agents and reviewed by a human before opening. How we work: https://github.com/paper-rerun/paper-rerun/blob/main/METHOD.md

[codacy-production]

Up to standards ✅

🟢 Issues 2 high

Results:
2 new issues

Category	Results
Security	2 high

View in Codacy

🟢 Metrics 0 complexity

Metric	Results
Complexity	0

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}