Homework 4 (2026) Q5 std mismatch — 0.013 instead of 0.007 (solver must be liblinear) #482

Course

machine-learning-zoomcamp

Question

Why does my Homework 4 (2026) Q5 cross-validation std not match any option — I get 0.013 instead of 0.007?

Answer

Fix first — copy the model constructor exactly as the homework specifies:

KFold(n_splits=5,shuffle=True, random_state=1)
# per fold, with a fresh DictVectorizer fit on the train fold only:
model = LogisticRegression(solver='liblinear', C=1.0, max_iter=1000)

The bare constructor LogisticRegression(max_iter=1000) silently uses the default solver lbfgs. On the 2026 lead-scoring data it throws ConvergenceWarning: lbfgs failed to converge and produces fold AUCs ~[0.776, 0.787, 0.799, 0.811, 0.780] with std 0.013 — a different multiple-choice option. With solver='liblinear' the folds are ~[0.731, 0.730, 0.736, 0.741, 0.721] with std 0.007, the correct answer. Same data, same seed, wrong solver, wrong answer.

Why this happens:

lbfgs and liblinear optimize differently on unscaled one-hot features, so they converge to different weights and therefore different per-fold AUCs. The warning in your output is the tell: if you see ConvergenceWarning in Q5, your solver is wrong. Related trap in the same notebook: Q1 must exclude converted itself from the single-variable AUC loop (it scores 1.000 against itself and answers nothing).

Canonical homework (pinned 2026 release):

https://github.com/DataTalksClub/machine-learning-zoomcamp/blob/main/cohorts/2026/homework/04-evaluation/homework.md

Related lesson: 04-evaluation/07-cross-validation (mean = performance, std = stability across folds).

Checklist

• I have searched existing FAQs and this question is not already answered
• The answer provides accurate, helpful information
• I have included any relevant code examples or links