Use Case: Partition Receiving Entire Remaining Feature Count

Requirement: Distribute all counts of a feature between three partitions as equally as possible, with none in the default partition.

If you declare three partitions, two with a count specifier of 33% and one with 34%, integer rounding might cause some counts to slip into the default partition. You can avoid this by replacing the last count specifier with the remainder keyword, which will match all available remaining counts.

Note that any partition lower than the one specifying the remainder keyword in the model definition receives zero counts for that feature.

Model Definition Example

model "exampleModel" {

partitions {

partition "p1" {                   // Total feature count of f1: 10

"f1" 1.0 33%                   // Feature count in p1: 3

}

partition "p2" {

"f2" 1.0 33%                   // Feature count in p2: 3

}

partition "p3" {

"f3" 1.0 remainder             // Feature count in p3: 4

}

}