Principle:Interpretml Interpret Optimal Transport Selection
Metadata
| Field | Value |
|---|---|
| Sources | Interpretml_Interpret |
| Domains | Machine_Learning, Data_Selection |
| Updated | 2026-02-07 |
Overview
The SPOTgreedy algorithm selects a representative subset of prototypes from a dataset using optimal transport theory for data summarization and instance selection.
Description
SPOT_GreedySubsetSelection implements the SPOTgreedy algorithm from the SPOT (Scalable Prototype Optimal Transport) framework. It selects a set of representative prototypes from a dataset by greedily minimizing an optimal transport cost. The selected prototypes form a compact summary of the full dataset, useful for data summarization, prototype-based explanation, and efficient model training. The algorithm operates on pre-computed distance matrices and iteratively selects points that best represent the overall data distribution.
Usage
Use optimal transport selection when you need to select a representative subset of data points from a large dataset, for example as prototypical examples for model explanation or as a compressed training set.