Sum models

Purpose

Blend trees and counters of two or more trained CatBoost models into a new model. Leaf values can be individually weighted for each input model. For example, it may be useful to blend models trained on different validation datasets.

Execution format

catboost model-sum -m [<model with 1.0 weight>] [--model-with-weight <path_to_model>=<weight>] -o <result model path>

Options

OptionDescriptionDefault value
-m

The path to the model that should be blended with the others. The weight of the leafs of this model is set to 1.

This parameter can be specified several times to set the required number of input models.

At least one of these parameters should be set at least once

--model-with-weight

The path to the model that should be blended with the others. Use this parameter to set the individual weight for the leaf values of this model. Use the equal sign as the separator.

This parameter can be specified several times to set the required number of input models.

-o

The path to the output model obtained as the result of blending the input ones.

Required parameter
--ctr-merge-policyThe counters merging policy. Possible values:
  • FailIfCtrsIntersects — Ensure that the models have zero intersecting counters.
  • LeaveMostDiversifiedTable — Use the most diversified counters by the count of unique hash values.
  • IntersectingCountersAverage — Use the average ctr counter values in the intersecting bins.
IntersectingCountersAverage