model_unified.object: Unified model representation

In ModelOriented/treeshap: Compute SHAP Values for Your Tree-Based Models Using the 'TreeSHAP' Algorithm

Unified model representation

Description

model_unified object produced by *.unify or unify function.

Value

List consisting of two elements:

model - A data.frame representing model with following columns:

Tree	0-indexed ID of a tree
Node	0-indexed ID of a node in a tree. In a tree the root always has ID 0
Feature	In case of an internal node - name of a feature to split on. Otherwise - NA
Decision.type	A factor with two levels: "<" and "<=". In case of an internal node - predicate used for splitting observations. Otherwise - NA
Split	For internal nodes threshold used for splitting observations. All observations that satisfy the predicate Decision.type(Split) ('< Split' / '<= Split') are proceeded to the node marked as 'Yes'. Otherwise to the 'No' node. For leaves - NA
Yes	Index of a row containing a child Node. Thanks to explicit indicating the row it is much faster to move between nodes
No	Index of a row containing a child Node
Missing	Index of a row containing a child Node where are proceeded all observations with no value of the dividing feature
Prediction	For leaves: Value of prediction in the leaf. For internal nodes: NA
Cover	Number of observations seen by the internal node or collected by the leaf for the reference dataset

data - Dataset used as a reference for calculating SHAP values. A dataset passed to the *.unify, unify or set_reference_dataset function with data argument. A data.frame.

Object has two also attributes set:

model	A string. By what package the model was produced.
missing_support	A boolean. Whether the model allows missing values to be present in explained dataset.

See Also

unify

ModelOriented/treeshap documentation built on Jan. 27, 2024, 10:57 p.m.