loss - Kumo.ai

`loss: <list[str]>` (Optional)

Description

The loss type to use during model optimization depending on the task type. Available Options:

Task Type	Available loss options
Binary Classification/Multilabel Classification	`binary_cross_entropy` (default) `focal`
Multiclass Classification	`cross_entropy` (default) `ordinal_log`
Regression/Forecasting	`mae` `mse` `huber` (default) `wape` `multi_quantile`
Temporal Link Prediction	`cross_entropy` (default) `binary_cross_entropy`
Static Link Prediction	`cross_entropy` (default) `binary_cross_entropy`
Multilabel Ranking	`cross_entropy` (default) `ordinal_log`

By default, focal loss uses an alpha value of 0.25 (the weighting factor to balance positive vs. negative examples), and a gamma value of 2.0 (the balance between easy vs. hard examples). You can further customize this in the model plan by replacing the string by a dictionary:

loss:
- name: focal
  alpha: 0.5
  gamma: 4.0

By default, huber loss uses a delta value of 1.0. You can further customize this in the model plan by replacing the string by a dictionary:

loss:
- name: huber
  delta: 2.0

Use multi_quantile for regression or forecasting tasks when you want prediction intervals in addition to the median prediction. It trains multiple quantiles with pinball loss and writes TARGET_PRED together with 27 quantile columns named q_0.005, q_0.01, …, q_0.995:

loss:
- multi_quantile

The full quantile column set is:

q_0.005 q_0.01  q_0.02  q_0.025 q_0.05  q_0.1   q_0.15
q_0.2   q_0.25  q_0.3   q_0.35  q_0.4   q_0.45  q_0.5
q_0.55  q_0.6   q_0.65  q_0.7   q_0.75  q_0.8   q_0.85
q_0.9   q_0.95  q_0.975 q_0.98  q_0.99  q_0.995

Supported Task Types

All