lesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept driftlesson 2 of 6 · data drift and concept drift
Write drift_check(train_stats, window, threshold=2.0) returning [(feature, z)] for every feature whose serving-window mean shifted more than the threshold (z rounded to 1). Run it over the three-feature window, print the flagged list and the closing line "investigate before retraining - a drift flag is a lead, not a verdict".