weight_mode: (sample (default) | weighted_loss | mix) (Optional)

Description

weight_mode specifies how Kumo handles imbalanced datasets.
  • sample: samples training examples with replacement according to the label weight distribution.
  • weighted_loss: weighs training examples in the loss function according to the label weight distribution.
  • mix: samples training examples with replacement according to the label weight distribution, but will re-weigh them in the loss function according to the inverse of the given weight.
This option is applicable only when majority_sampling_ratio is specified.

Supported Task Types

  • Binary Classification

Default Values

Run ModeDefault Value
FASTsample
NORMALsample
BESTsample