Skip to learning content

DATA SCIENCE PYTHON PLAYGROUND

Machine Learning · Learn / Refresh

← Choose and Explain Models lessonsQUESTIONS · MODELS · EVIDENCE
Communicate evidence · ML-M-K1 · 25–35 MIN

Choose and Explain Models checkpoint

Retrieval 1

Exercises within this concept

  1. Retrieval 1Retrieve and applyCurrent exercise
Retrieval 1 · LINE24

Audit these five cases and write a concise report: (1) Wine candidate A/B CV RMSE 0.8/0.81 with fold variation 0.1; (2) Penguin classifier accuracy .95 but minority recall .2; (3) K-Means k=3/4 silhouettes .51/.52 with very different sizes; (4) Ward profiles computed on a 500-row sample but claimed for 569 rows; (5) PCA retains 90% in seven axes but displays only two.

For each, state the task, defensible conclusion and a limitation.

Audit task-family choice and evidence across the complete curriculum.

This practice: Audit task-family choice and evidence across the complete curriculum. Next: Apply the workflow in an independent challenge.

Supporting concepts: Explain a result responsibly →

Remember the idea

Use the supplied briefs to distinguish predictive comparison, grouping and representation. Judge evidence within each task rather than ranking incompatible metrics.

Audit task-family choice and evidence across the complete curriculum.QuestionPredict quantityPredict labelDiscover groupsReduce dimensions
Schematic · Audit task-family choice and evidence across the complete curriculum.Scroll the diagram horizontally if needed.
Hint 1 — Think

Reconstruct the evidence boundary appropriate to each task before judging its result.

Hint 2 — Tools

Validation variation, minority errors, cluster profiles, sample scope and retained variance.

Hint 3 — Approach

For each case name the task, connect the stated evidence to a defensible conclusion and identify the missing or limited claim.

Explained solution

Wine: small CV difference does not establish universal superiority; consider uncertainty and cost. Penguins: accuracy hides minority errors; inspect macro F1 and per-class evidence. K-Means: either k may be defensible if supported by purpose and profiles. Ward: labels and profiles describe sampled rows only. PCA: the 2D view is distinct from the seven-axis retained representation. None of these establishes causal effects.

Wine: small CV difference does not establish universal superiority; consider uncertainty and cost. Penguins: accuracy hides minority errors; inspect macro F1 and per-class evidence. K-Means: either k may be defensible if supported by purpose and profiles. Ward: labels and profiles describe sampled rows only. PCA: the 2D view is distinct from the seven-axis retained representation. None of these establishes causal effects.

Helpful prior knowledge: Model-choice retrieval These links are guidance, not locks.

Sources and API context

Examples run with this Playground’s scikit-learn 1.4.2 / Pyodide 0.26.4 runtime.

Use the evidence and state a limitation. This is self-review, not a keyword test.