get_object_importance

Calculate the effect of objects from the train dataset on the optimized metric values for the objects from the input dataset:

  • Positive values reflect that the optimized metric increases.
  • Negative values reflect that the optimized metric decreases.

The higher the deviation from 0, the bigger the impact that an object has on the optimized metric.

The method is an implementation of the approach described in the Finding Influential Training Samples for Gradient Boosted Decision Trees paper .

Currently, object importance is supported only for the following loss functions.

Logloss

CrossEntropy

RMSE

MAE

Quantile

Expectile

LogLinQuantile

MAPE

Poisson

Method call format

get_object_importance(pool,
                      train_pool,
                      top_size=-1,
                      type='Average',
                      update_method='SinglePoint',
                      importance_values_sign='All',
                      thread_count=-1,
                      verbose=False,
                      log_cout=sys.stdout,
                      log_cerr=sys.stderr)

Parameters

pool

Description

The data for calculating object importances.

Possible types

catboost.Pool

Default value

Required parameter

train_pool

Description

The dataset used for training.

Possible types

catboost.Pool

Default value

Required parameter

top_size

Description

Defines the number of most important objects from the training dataset. The number of returned objects is limited to this number.

Possible types

int

Default value

-1 (top size is not limited)

type

Description

The method for calculating the object importances.

Possible values:

  • Average — The average of scores of objects from the training dataset for every object from the input dataset.
  • PerObject — The scores of each object from the training dataset for each object from the input dataset.

Possible types

string

Default value

Average

update_method

Description

The algorithm accuracy method.

Possible values:

  • SinglePoint — The fastest and least accurate method.
  • TopKLeaves — Specify the number of leaves. The higher the value, the more accurate and the slower the calculation.
  • AllPoints — The slowest and most accurate method.

Supported parameters:

For example, the following value sets the method to TopKLeaves and limits the number of leaves to 3:

TopKLeaves:top=3

Possible types

string

Default value

SinglePoint

importance_values_sign

Description

Defines the type of effect that the objects from the training dataset must have on the optimized metric value for objects from the input dataset. Only the appropriate objects are output.
Possible values:

  • Positive
  • Negative
  • All

Possible types

string

Default value

All

thread_count

Description

The number of threads to use for operation.

Optimizes the speed of execution. This parameter doesn't affect results.

Possible types

int

Default value

-1 (the number of threads is equal to the number of processor cores)

log_cout

Output stream or callback for logging.

Possible types

  • callable Python object
  • python object providing the write() method

Default value

sys.stdout

log_cerr

Error stream or callback for logging.

Possible types

  • callable Python object
  • python object providing the write() method

Default value

sys.stderr

Type of return value

Two lists of lists with indices and scores.

For example, if the input dataset contains 3 rows and the training dataset contains 4 rows, the indices list takes the following structure:

[[1, 2, 3, 4], [4, 3, 2, 1], [1, 3, 4, 2]]

The scores list has the same structure with the corresponding scores instead of indices.