Instance Space Analysis Toolkit v0.9.0 — July 2026

Version 0.9.0 is a complete overhaul of the MATLAB ISA Toolkit, the computational engine behind the University of Melbourne’s MATILDA platform for algorithm benchmarking and selection. This release brings the codebase in full alignment with two canonical publications — Smith-Miles & Muñoz (2023, ACM CSUR) and Simpson et al. (2025, Machine Learning) — and introduces substantial new functionality alongside a comprehensive correctness pass.

The centrepiece of this release is a new InstanceSpace class-based API that replaces the previous script-driven workflow. Users can now run the full analysis pipeline — or any subset of its stages — in a single, reproducible object, with options adjustable between stages and automatic invalidation of stale downstream results. The existing buildIS/exploreIS entry points are retained for backward compatibility.

Key new capabilities include native support for three-dimensional instance spaces, automatic viewpoint optimisation for 3D projections, and a Partial Least Squares projection alternative for non-full-rank datasets. Algorithm selection has been modernised with a configurable classifier registry — supporting k-NN, SVM, decision trees, Naïve Bayes, linear and ensemble methods — that replaces the deprecated LIBSVM dependency. A migration utility (ISAmigrateModel) allows users to migrate pre-trained models to the new format without rerunning the full pipeline.

The release also includes a new AI-assisted analysis capability via a bundled Claude Code skill, enabling an AI assistant to help run, interpret, and debug ISA pipeline runs directly within the repository.

This release targets MATLAB R2025a or later. The licence has changed from GPL v3 to the PolyForm Noncommercial License 1.0.0, permitting non-commercial research and educational use. Commercial use requires prior written permission from the copyright holders.

The toolkit is available at github.com/andremun/InstanceSpace and on MATLAB Central File Exchange.