Simple Weak Coresets for Non-Decomposable Classification Measures
This work addresses the challenge of efficient coreset construction for specific classification metrics, but it is incremental as it extends existing coreset techniques to a new supervised setting.
The paper tackles the problem of constructing coresets for supervised classification with non-decomposable evaluation measures like F1 score and Matthews Correlation Coefficient, showing that stratified uniform sampling achieves theoretical lower bounds and empirical performance comparable to more complex methods.
While coresets have been growing in terms of their application, barring few exceptions, they have mostly been limited to unsupervised settings. We consider supervised classification problems, and non-decomposable evaluation measures in such settings. We show that stratified uniform sampling based coresets have excellent empirical performance that are backed by theoretical guarantees too. We focus on the F1 score and Matthews Correlation Coefficient, two widely used non-decomposable objective functions that are nontrivial to optimize for and show that uniform coresets attain a lower bound for coreset size, and have good empirical performance, comparable with ``smarter'' coreset construction strategies.