Short ordered sequences through the labs, with progress read live from your scored runs.
0 / 8 labs passed
Fit a straight line, bend it into a curve, then keep the fit steady when corrupt rows try to drag it off course.
Closed-form least squares on clean data is the gentlest start, feature engineering for the quadratic comes next, and outlier-robust refitting is the hardest of the three.
Learn the logistic core on easy data, then carry it through keyword counts and a lopsided default rate.
A balanced two-feature classifier teaches the boundary first; F1 on spam and credit data only lands once the logistic core is second nature.
Find structure without labels, then engineer the features a linear model cannot invent on its own.
Clustering shows what unlabeled structure buys you, and XOR closes the arc by showing why the right representation beats a fancier model.
Four labs where the obvious baseline looks reasonable and still misses the metric.
Ordered from the loudest failure to the quietest: majority-class guesses score zero F1, then thresholds, then outliers, then a boundary no straight line can draw.