1
Lock the baseline
Define a deterministic or majority baseline before candidate comparison.
Required evidence: Baseline predictions and confusion counts.
0/80 minimum characters
Builder mini-project·90-120 minutes
Classical Model Selection Project
Compare a linear rule, a decision stump, and a nearest-centroid classifier under one shared evaluation contract.
Scenario
A small operations team needs a transparent binary classifier and cannot support a large serving stack. Your recommendation must balance error cost, stability, and explainability.
You will demonstrate
- Use identical splits and metrics for every candidate.
- Diagnose model-specific failure modes.
- Choose complexity only when evidence justifies it.
Project evidence
Show the work, not a checked box.
Each response is stored in this browser as you type. Include metrics, test output, or a decision rationale wherever the deliverable asks for it.
2
Compare candidates
Evaluate all candidates on identical held-out rows and threshold policy.
Required evidence: One scorecard with error counts and cost.
0/80 minimum characters
3
Stress the decision
Perturb one feature and identify which model changes predictions unexpectedly.
Required evidence: A small sensitivity table.
0/80 minimum characters
4
Defend the selection
Recommend a model or the baseline and name a rollback trigger.
Required evidence: Decision memo tied to the scorecard.
0/80 minimum characters
Runnable Python lab
Select by cost, then simplicity
Compute cost from fixed confusion counts and break a tie in favor of fewer fitted parameters.
Project defense
Why must every candidate use the same held-out rows?
Rubric self-review
Rate the evidence, not your effort: 0 missing, 1 weak, 2 adequate, 3 strong. All criteria must be reviewed, but a low honest score does not get hidden.
Comparisons use identical evidence.
Complexity is treated as a cost.
Failure behavior is demonstrated, not merely named.
The recommendation is reversible.
Useful references
Project completion gate
Completion is controlled by stored evidence, deterministic tests, a decision defense, rubric review, and the artifact when required.
Evidence pendingCode pendingDefense pendingRubric pending
Optional cloud portfolio
Submit evidence across devices.
An account is required. Submit only when the local completion gate passes. AI review is advisory and separate from deterministic completion.