CLASSICAL MACHINE LEARNING · PROJECT

A real-data model comparison.

Compare a simple reference with a fitted candidate, defend the split and examine errors before making a claim.

Work independently or revisit validation design, linear regression and residual boosting. No earlier completion is required. Learner pilots and human artifact review remain pending.

Compare a real-data prediction

Predict petal length from sepal length, both in centimetres, in the small Iris dataset. Species is used for partition balance and error inspection, never as a model input. Start with a mean prediction; a more complicated model does not automatically win.

Public Iris measurements have been used in preceding lessons. Reserved within this workflow does not mean previously unseen or independently authenticated. Your report, objective transfer answers and pending human review have separate statuses. Everything is saved locally in version 4 progress; no account or cloud sync is implied.

Inspect dataset provenance and split

Fisher, R. (1936). Iris [Dataset]. UCI Machine Learning Repository. · CC BY 4.0. Dataset source · License · Download local CSV.

Deduplicate identical original four-feature vectors; within each species in first-occurrence order, group index modulo five: 0 evaluation, 1 development, 2–4 training. Predictor: sepal length; target: petal length, both cm. Species stratifies the split and supplies error slices, not a model input.

Local CSV: 5977 bytes. SHA-256: b34e5958ea7f7625f6a43c060b28d9848e6b810c5d09652b2335ef4619db9df8. The original asset has 150 rows; identical measurement vectors reduce this task to 149 groups. No synthetic missingness or unit alteration is applied here. The asset’s feasibility manifest describes a different development split; this project uses iris-petal-from-sepal-grouped-v1.

Training
89 groups
Fit each model

Development
30 groups
Compare and select

Frozen evaluation
30 groups
Reveal explicitly

Fit, compare and freeze

Model family

Worker unavailable? Direct controls support the full mean-versus-line project with bounded CPU arithmetic. Ensembles run only in workers to keep the page responsive.

0/12 retained comparisons; known comparison count 0. Repeated settings are repeated development work, not new independent evidence. Fitting runs in a local CPU worker with at most forty stages per candidate. Cancellation, timeout, hiding, leaving the viewport or navigation retain the preceding saved state.

Calculate a baseline, then compare a line or ensemble.

Report evidence for your chosen path Project variant Prediction task and intended use

Predict petal length from sepal length for a classroom collection. State available inputs, units, acceptance criteria and excluded uses.

Data card and split

Cite source/license, 149 measurement groups, 89/30/30 partition, duplicate handling and prior exposure. Explain why these are not verified specimen identities.

Comparison and selection

Compare the mean baseline with a fitted line or ensemble using development MAE/MSE. State your comparison allowance and why you selected this model.

Error analysis

Cite a specific frozen-evaluation error and species slice. Explain a plausible limitation without asserting a cause that the measurements cannot establish.

Limits and new evidence

Separate this public-data frozen comparison from unseen transfer. State missing populations, uncertainty and the next independently collected evidence.

Recommendation and reproduction

Guided: explain your experiment. Implementation: provide run instructions and tests. Decision: recommend proceed, defer or no model against the stated criteria and operating constraints.

Report: draft. Submission requires a baseline and another candidate, frozen evaluation and at least twenty characters in each field (plus implementation text for that variant). This only checks presence. Human artifact review: pending. Download a My Work backup to preserve editable state.

Separate transfer evidence

Try a different problem independently

These two fixed cases use different data and separate scoring arithmetic from the guided Iris project. Work without the walkthrough first. Opening and submissions are saved; the first submission is retained, with at most two practice retries. Reopening, resetting or importing an earlier backup does not erase known exposure. This is a local objective check, not authenticated exam supervision or human review of your implementation.

0/2 cases passed on first submission. Latest/practice passes: 0/2. Project submission and these outcomes are separate.

Delivery-demand validation

A concentrated-error release decision

Interpret the comparison

Training fits each model; development informs selection; the explicit frozen evaluation describes the selected recipe on a reserved partition of previously exposed public data. MAE is measured in centimetres and MSE in squared centimetres. Inspect individual errors before making a claim.

Read the complete explanation and worked error calculations

Back up your saved project in My Work · Explore the expanded curriculum