Los enunciados están en inglés a propósito, porque la prueba lo estará. La semilla sale de la fecha:
el mismo proyecto mañana tiene las mismas preguntas sobre otros números.
Telco subscriptions
The default shape of the assessment: text-typed numerics, duplicates, a sentinel, and one leaking column.
La lección: A blank string types a numeric column as object; a sentinel is not a number; one column would not exist at scoring time
90 min7 tareas
100 ptsROC-AUC
Sin intentos en este navegador
Retail forecasting
Regression on a panel where the validation split is the trap and the holdout is the future.
La lección: Two date formats in one column, and a validation split that must respect time
90 min6 tareas
100 ptsRMSE
Sin intentos en este navegador
Consumer credit
Five per cent positives and a category the scoring file invented: imbalance and encoder alignment.
La lección: 5% positives, so accuracy is worthless; and a category the scoring file invented
90 min5 tareas
100 ptsAverage precision + recall in a 5% budget
Sin intentos en este navegador
Manufacturing quality
Wide sensor table, four classes, and a rare one that accuracy would let you ignore.
La lección: Dead and redundant columns, mixed units, near-duplicate rows, a rare class that matters
90 min6 tareas
100 ptsMacro F1
Sin intentos en este navegador
Hospital operations
Several rows per patient: the split is the exam question, and a random one passes it wrongly.
La lección: Several rows per patient: split on the group or validate against yourself
90 min5 tareas
100 ptsROC-AUC, patients disjoint
Sin intentos en este navegador