THE GARDEN

Evaluation and analysis

Updated 06 Oct 2026

• •

  Analysis and evaluation

  • •

    Map each question to data, preprocessing, analysis, assumptions, otuputs, and interpretation criteria; distinguish planned analyses from exploration

  • •

    training data fit a model, validation data guide choices, and test data estimate final performance after choices are fixed. split according to the intended claim: keep related or duplicate cases together and preserve time order for future prediction

• •

  Cross-validation and leakage

  • •

    in $k$-fold cross-validation, train on $k-1$ folds and test on the remaining fold in turn. Stratified, grouped, or time-aware folds address class balance, dependence, or time order.

  • •

    Fit learned preprocessing inside each training fold. Leakage occurs wheninformation unavailable at prediction time affects evaluation, such as scaling before splitting, duplicates across partitions, futrue data, or repeated test-set tuning

• •

  Confidence interval

  • •

    convidence interval covers the target parameter in about $95\%$ of repeated samples under its assumptions; it does not mean $95\%$ of observations lie in the interval or remove bias.

  • •

    The handout poll example ($n=1{,}082$, $\hat{p}=0.65$) gives a simple-random-sample approximation of $65\%\pm2.8$ percentage points, which does not validate its design, weighting, or measurement.

  • •
• •

◌ Explore connections in Graph view