Fix #116 and #129: label import bug and matplotlib backend conflict

[jbohnslav]

Fixes

Fix #116: add_label_to_project() silently deletes first behavior column

The old code used pd.read_csv(path, index_col=0) which assumes every CSV has a DEG-style unnamed numeric index column. External CSVs without one had their first data column silently consumed as the index.

Fix: Read without index_col and explicitly detect/drop unnamed index columns.

Tests added: Three new test cases covering DEG-style CSVs, external CSVs without index, and external CSVs with background but no index.

Fix #129: matplotlib backend conflict when importing training modules

plt.switch_backend('agg') was called at module level in flow_generator/train.py, feature_extractor/train.py, and sequence/train.py. This killed interactive matplotlib for anyone importing these modules in notebooks or scripts.

Fix: Moved the call inside the training functions. Training still uses the agg backend; importing the module no longer overrides the user's backend.

Testing

All 28 tests pass (pytest -v -m 'not gpu'), including 3 new tests for #116.

[claude]

Claude finished @jbohnslav's task in 1m 38s —— View job

---

Code Review

• Gather context and understand the PR
• Review Fix projects.add_label_to_project is broken #116: label import changes in projects.py and tests
• Review Fix Error with matplotlib backend while importing train files  #129: matplotlib backend changes in train files
• Review other changed files (pyproject.toml, uv.lock)
• Post final review

---

Both fixes are clean and address the root causes. A few observations below.

Fix #116 — add_label_to_project() (deepethogram/projects.py:174-178)

The logic is correct. When pandas reads a CSV with an empty header cell (DEG-generated style: ,background,behavior1,...), it names that column "Unnamed: 0", so str(first_col).startswith("Unnamed") reliably catches it.

Minor: the first_col == "" check is unreachable in practice. Pandas normalizes empty column headers to "Unnamed: X", so this branch will never be true for CSV files. It's harmless dead code, but may be confusing to future readers.

# first_col == "" will never match; pandas always produces "Unnamed: X"
if first_col == "" or str(first_col).startswith("Unnamed"):

Fix this →

Re: Codex concern about named index columns — jbohnslav's response is correct. DEG always saves with df.to_csv(label_dst) (no index_label), producing an unnamed index, so any re-read DEG file gets correctly detected and dropped. Named index columns (e.g. "frame") from other tools would not match the heuristic and would be kept as a behavior column, which is the right behavior here.

Tests are thorough. The three test cases cover all important paths:

• DEG-style CSV with unnamed numeric index ✓
• External CSV with no index and no background ✓ (the actual regression case for projects.add_label_to_project is broken #116)
• External CSV with background but no index ✓

Fix #129 — matplotlib backend (feature_extractor/train.py, flow_generator/train.py, sequence/train.py)

Correct fix. Moving plt.switch_backend("agg") from module level to the first line of the training function means importing these modules no longer clobbers the active backend (fixing notebook/interactive use), while training runs still use agg.

One thing worth being aware of: if a caller sets up a non-agg backend before calling a training function, it will be silently overridden. This is intentional (training shouldn't try to display windows), but it might be worth a comment so it's not mistakenly "fixed" in the future:

# Use non-interactive backend for training; don't move this to module level
plt.switch_backend("agg")

This is entirely a style suggestion, not a bug.

pyproject.toml

Version bump 0.3.0 → 0.4.0 is appropriate given the fixes address data-correctness and API-behavior bugs.

Overall

The PR is well-scoped, the fixes address root causes rather than symptoms, the tests clearly document the regression, and the changes are minimal. No blocking issues.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a1d95860d8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".