Bugl — code interview practice that feels like a game

Loading lesson path

Model lab

Choose the feature before trusting the prediction

Build a tiny prediction from one feature and keep a holdout set for checking it.

1/3 checks

Puzzle target

Use satisfaction as the feature and keep 75% of rows for training.

○Model feature is satisfaction

○Training split is 75%

✓Filtered data still has a holdout row

Working dataset

City	Region	Visits	Signups	Revenue	Satisfaction
Vancouver	West	1,200	156	$18,400	86
Calgary	West	860	95	$9,900	74
Toronto	Central	2,100	252	$32,600	82
Montreal	East	1,580	181	$21,200	79
Halifax	East	640	83	$8,700	88

Revenue visual

Vancouver

$18,400

Calgary

$9,900

Toronto

$32,600

Montreal

$21,200

Halifax

$8,700

Tiny model

Feature

Visits

Train

60%

Predicted revenue

$21,793

This is intentionally small: change one feature, keep a holdout split, and explain what changed before trusting the model.

Work directly with the dataset lab below. The controls change the rows, derived columns, visual, and tiny model summary in place.

Practice surface	What you manipulate
Dataset	Campaign rows with visits, signups, revenue, satisfaction, and promo cost
Transform	Filters, derived net revenue, metric choice, and sorting
Model	One-feature prediction with a train and holdout split
Goal	Choose one feature, reserve holdout rows, and inspect the prediction.

Use satisfaction as the model feature and keep a 75% training split.
Watch the checklist in the lab update as the dataset state changes.
Use the table, visual, and model card together before deciding what the data says.