[buyan-kh]

Summary

• check_intervals now validates chrom / start / end by column name (order-independent), so grelu.sequence.utils.resize and other callers no longer fail when those columns are present but reordered.
• Updated resize / format docstrings to document name-based interval columns; DFSeqDataset also selects interval columns by name for consistency.
• Added unit tests for out-of-order (and extra) interval columns in get_input_type and resize.

Fixes #34

Test plan

• pytest tests/test_sequence.py::test_get_input_type tests/test_sequence.py::test_resize
• Confirm tutorial-style dataframe with columns like name, end, chrom, start works with resize

[avantikalal]

Thanks for this — the root cause tracking is spot on. I confirmed check_intervals()'s positional column check (df.columns[:3] == ["chrom","start","end"]) and DFSeqDataset.__init__'s df.iloc[:, :3] / df.iloc[:, 3:] were the only two places in the codebase making that positional assumption (grepped for columns[:3] / iloc[:, :3] repo-wide), so this fix is complete in scope — no other callers need touching. resize() itself already indexed interval columns by name internally, so that diff is correctly docs/type-hints only.

Two things I'd like addressed before merging:

1. Stale docstring in DFSeqDataset (src/grelu/data/dataset.py) — the class docstring still says "genomic intervals in the first 3 columns," which is no longer accurate now that column order doesn't matter. Worth updating alongside the resize()/format.py docstring changes you already made.

2. Missing test coverage for the dataset.py change itself — the new tests (test_get_input_type, test_resize in tests/test_sequence.py) cover check_intervals/resize, but tests/test_dataset.py already has extensive DFSeqDataset coverage and gets no new case for reordered interval columns. Since DFSeqDataset.__init__'s seqs/labels split is one of the two files this PR actually changes, I'd like a test that constructs a DFSeqDataset from a dataframe with out-of-order chrom/start/end (plus label columns) and asserts the labels extracted are correct — that's the part most likely to silently regress later.

Logic itself looks correct and the existing tests are good (nice touch testing extra columns too). Requesting changes mainly for the test gap above.