vimpTable-class: Variable importance table

In familiar: End-to-End Automated Machine Learning and Model Evaluation

Variable importance table

Description

A vimpTable object contains information concerning variable importance of one or more features. These objects are created during feature selection.

Details

vimpTable objects exists in various states. These states are generally incremental, i.e. one cannot turn a declustered table into the initial version. Some methods such as aggregation internally do some state reshuffling.

This object replaces the ad-hoc lists with information that were used in versions prior to familiar 1.2.0.

Slots

vimp_table

Table containing features with corresponding scores.

vimp_method

Method used to compute variable importance scores for each feature.

run_table

Run table for the data used to compute variable importances from. Used internally.

score_aggregation

Method used to aggregate the score of contrasts for each categorical feature, if any,

encoding_table

Table used to relate categorical features to their contrasts, if any. Not used for all variable importance methods.

cluster_table

Table used to relate original features with features after clustering. Variable importance is determined after feature processing, which includes clustering.

invert

Determines whether increasing score corresponds to increasing (FALSE) or decreasing rank (TRUE). Used internally to determine how ranks should be formed.

project_id

Identifier of the project that generated the vimpTable object.

familiar_version

Version of the familiar package used to create this table.

state

State of the variable importance table. The object can have the following states:

• initial: initial state, directly after the variable importance table is filled.

• decoded: depending on the variable importance method, the initial variable importance table may contain the scores of individual contrasts for categorical variables. When decoded, data in the encoding_table attribute has been used to aggregate scores from all contrasts into a single score for each feature.

• declustered: variable importance is determined from fully processed features, which includes clustering. This means that a single feature in the variable importance table may represent multiple original features. When a variable importance table has been declustered, all clusters have been turned into their constituent features.

• reclustered: When the table is reclustered, features are replaced by their respective clusters. This is actually used when updating the cluster table to ensure it fits to a local context. This prevents issues when attempting to aggregate or apply variable importance tables in data with different feature preprocessing, and as a result, different clusters.

• ranked: The scores have been used to create ranks, with lower ranks indicating better features.

• aggregated: Score and ranks from multiple variable importance tables were aggregated.

See Also

get_vimp_table, aggregate_vimp_table

familiar documentation built on Sept. 30, 2024, 9:18 a.m.