0.2.0

TL013 and TL014, both from @jay23606. TL013 flags data rescaled with its own mean, std, min or max before a split. TL014 flags a training DataLoader that never shuffles. The regression corpus is now 20 buggy scripts. Full list: https://github.com/Satvik-jain/trainlint/blob/main/CHANGELOG.md

trainlint 0.1.1

README links and the demo image are absolute, so they render on the PyPI page. No code changes since 0.1.0.

trainlint 0.1.0

First release.

trainlint checks Python training and evaluation code for ML bugs that do not crash, and ships
agent skills that steer coding agents away from the same bugs.

• 12 rules, TL001 to TL012: leakage before splits, eval mode and no_grad, seeding, mixed
  precision, checkpoints, loss inputs, gradient clearing, tensor accumulation, metric misuse and
  evaluation loaders. Examples for each are in docs/rules.md.
• trainlint check with text and JSON output, --select, --ignore, --exit-zero,
  --color, and # trainlint: ignore[RULE] comments.
• Four Agent Skills and trainlint skills install.
• A regression corpus of 18 buggy scripts with fixed twins (python -m bench) and a spot check on
  93 files of pytorch/examples, described in the README.

It matches names and reads one file at a time, so expect some false positives. Please report them.