[solarsys]

Summary

examples/tutorials/series still held the pre-release notebooks from #1281: they install PyHealth from GitHub and miss the fixes made since. This replaces them with the maintained Colab versions (the ones linked from pyhealth.dev and, after #1311, from the docs).

What changes:

• 00–08 replaced with the current Colab notebooks. These versions:
  • install pyhealth==2.1.0 from PyPI;
  • fix the label leaks found in review (Quickstart split after set_task, Tutorial 02 long-stay task);
  • describe the MIMIC mortality target accurately (death in hospital during the next admission);
  • embed a visual guide in each notebook and add a "Try it yourself" section.
• Tutorials 04 and 07 expanded, and both open on a free T4 GPU, with smaller fallback settings on CPU:
  • 04: all six models (LogisticRegression, MLP, RNN, Transformer, RETAIN, XGBoost) on one patient split, with each model's name on its training-log lines, commented code, and a guide to reading the log and results.
  • 07: label distributions for specialties and ICD codes, an explanation of the "UNEXPECTED weights" load report, and worked examples of correct predictions checked against simple baselines.
• Three new notebooks, whats_new_1/2/3_*.ipynb, cover splits and reproducibility, calibration and confidence intervals, and XGBoost and code mapping.
• README.md rewritten for the 12 notebooks.
• _source/ removed. The notebooks are now edited as notebooks. Running the old generator would overwrite them with the pre-2.1 versions, and nothing else in the repo uses it.

Outputs are kept (about 6 MB in total), so readers can see results on GitHub before running anything. Every notebook comes from a full Colab run with no errors in any cell; 04 and 07 ran on a T4.

Testing

• All 12 notebooks were run end to end on Colab (Run all). Every code cell executed, and no cell produced an error.
• Checked the files for credentials and personal paths; the only token mention is Hugging Face's standard "unauthenticated requests" warning.
• tools/check_pr_rules.py passes.

🤖 Generated with Claude Code

[github-actions]

Test Coverage

Total: 67.88% (16694 / 24594 lines)

No modified files under pyhealth/.

Workflow run · 670103b

[jhnwu3]

/review

[github-actions]

AI Review

suggestion: human review
decision: not available

This PR is too large for AI review (over 300 files or 200000 diff characters), so it is labeled status: manual and needs a human review.

[jhnwu3]

lgtm