You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Notebook fixes before the workshop: dropped CVD features, linear-only grid, notebook 02 solution crash #5
Hi @NajiaB, PR #4 was merged before review, so here are the notebook findings. I ran all 14 notebooks in a clean environment with the repo's requirements. The student workbooks run without problems. The issues below are in the solutions and the content.
1. Notebooks 04 and 05 drop half the CVD features. The ColumnTransformer lists only age, chol, and bp. Its default remainder="drop" silently discards bmi, smoker, and sex_male, and smoker has the largest single effect. Test ROC AUC on the notebook split:
Model
3 features
All 6 features
Logistic regression
0.624
0.673
Gradient boosting
0.715
0.724
2. The notebook 05 solution grid searches only activation="identity". This makes the MLP a linear model. With ReLU added to the grid, ReLU wins with a test AUC of 0.755, compared with 0.623. Notebook 04 asks when an MLP can beat logistic regression, but the solution grid rules out the answer.
3. The notebook 02 solution crashes. Cell 2 fails with NameError: name 'plt' is not defined. Cells 3 to 5 fail with NameError: name 'X_train' is not defined, because the omics train/test split is never created. Correction: the split is at the end of cell 2, so the plt crash stops it from running. Adding the matplotlib import fixes both errors.
4. The notebook 02 interpretation text needs edits.
It says "associated with SBP", but the target is inflammation_index.
It calls Lasso and Ridge "a scalable alternative to classical inference". Penalized coefficients do not give p-values or confidence intervals.
5. Smaller items:
The Colab workbooks load data from Jesse-Islam's personal repo. Please point them at QLS-MiCM/SupervisedML.
The README has a placeholder clone URL and refers to a nonexistent Exercises/notebooks folder. It also says pip install npm.
The notebook 04 hint says logistic regression is optimal for "truly linear" data. It should say linear log-odds.
Notebook 05 has the typos "Use use" and "GridSewarchCV".
Please remove .ipynb_checkpoints, __pycache__, and README_template.md.
Activity
NajiaB commentedon Oct 28, 2026
Hi Cesar,
Thanks a lot for this! I'll implement the fixes asap.
cmvcordova commentedon Oct 29, 2026
Fixed in #6.