Generalization in portfolio-based algorithm selection

TL;DR

This paper provides the first theoretical guarantees for portfolio-based algorithm selection, analyzing how training set size, portfolio complexity, and overfitting affect the performance of algorithm selectors.

Contribution

It introduces a comprehensive learning-theoretic analysis of portfolio construction and algorithm selection, highlighting the tradeoffs between portfolio size and overfitting risk.

Findings

Large portfolios lead to inevitable overfitting, even with simple selectors.

Increasing portfolio size improves potential coverage but raises overfitting risk.

Theoretical bounds relate training set size, portfolio complexity, and expected performance.

Abstract

Portfolio-based algorithm selection has seen tremendous practical success over the past two decades. This algorithm configuration procedure works by first selecting a portfolio of diverse algorithm parameter settings, and then, on a given problem instance, using an algorithm selector to choose a parameter setting from the portfolio with strong predicted performance. Oftentimes, both the portfolio and the algorithm selector are chosen using a training set of typical problem instances from the application domain at hand. In this paper, we provide the first provable guarantees for portfolio-based algorithm selection. We analyze how large the training set should be to ensure that the resulting algorithm selector's average performance over the training set is close to its future (expected) performance. This involves analyzing three key reasons why these two quantities may diverge: 1) the learning-theoretic complexity of the algorithm selector, 2) the size of the portfolio, and 3) the learning-theoretic complexity of the algorithm's performance as a function of its parameters. We introduce an end-to-end learning-theoretic analysis of the portfolio construction and algorithm selection together. We prove that if the portfolio is large, overfitting is inevitable, even with an extremely simple algorithm selector. With experiments, we illustrate a tradeoff exposed by our theoretical analysis: as we increase the portfolio size, we can hope to include a well-suited parameter setting for every possible problem instance, but it becomes impossible to avoid overfitting.