Update Signals example for v3 data

[ndharasz]

• Update the Signals example to download v3.0 train and live data.
• Keep the trained model versioned and use its training feature order for live predictions.
• Fix the local training entry point and add CI-backed example tests.

Validation: python -m unittest discover -s tests -v (2 passed), git diff --check, and confirmed both v3.0 datasets exist through SignalsAPI. Reviewed the full diff; no findings.

Docs: no change needed; the public Signals docs already describe v3.0 data.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-30T01:56:37.969070Z	cf39573	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[claude]

Looks good — a version bump plus a straightforward simplification (predict now uses the trained model's feature_name_ instead of re-deriving the training exclusion list), and train.py is fixed to match predict.py's actual API.

What was reviewed: the DATA_VERSION/TRAINED_MODEL_PREFIX bump and per-version model naming in predict.py; the switch from a manual feature_cols filter to model.feature_name_ in predict() — consistent since feature_name_ reflects whatever columns train() actually fit on; train.py's new call signature, which matches predict.train's real parameters (the old download_data/MODEL/predict_data_path references were already broken); and the new test's use of mocks (API, LGBMRegressor, joblib.dump) confirming no real network/training calls occur in CI.

Extended reasoning...

Change touches only the Signals example (predict.py, train.py), its new unit test, and a CI workflow addition — no auth, crypto, or permission-sensitive code. The core logic change (using model.feature_name_ instead of a manually filtered column list) is a safe simplification since feature_name_ always reflects the columns the model was actually fit on. The new test mocks the API/model/joblib so it doesn't hit the network, and the CI step only runs on one Python version matrix leg with the correct requirements file installed first. Small, self-contained, mechanical change with test coverage, so no human review is strictly necessary.