Introduced on day 35
Tuning budget
The number of configurations or trials (and the compute) a model was allowed to be tuned with; comparisons are fair only when it is comparable across models.
In detail
A network chosen from twelve trials set against a baseline run once at defaults compares effort, not model families. Report the number of trials for every model, spend the budget on training rows (cross-validation) or on a validation part, and score the chosen configurations once on a sealed test part.