Evolving Executable Pipeline Programs for AutoML with Language Models

2026-08-17

AI Summary

This study addresses the limitations of restricted search spaces and non-editable outputs in traditional AutoML by proposing LACE, a novel code-based tabular AutoML framework. Leveraging large language models as mutation operators within an evolutionary algorithm, LACE optimizes populations of scikit-learn-compatible pipelines at the code level. Unlike conventional approaches constrained by predefined structures, this paradigm generates transparent, editable code rather than opaque black-box models, thereby enhancing reusability and interpretability. Extensive evaluations across 68 OpenML tasks demonstrate that LACE significantly outperforms auto-sklearn and achieves performance comparable to AutoGluon. By unifying comprehensive search coverage with high transparency, LACE effectively overcomes the rigidity of existing AutoML systems while delivering human-readable solutions that facilitate downstream refinement and domain-specific adaptation.

Abstract

Automated machine learning (AutoML) systems search for pipelines within a space of preprocessing operators, learners, and hyper-parameters specified in advance: they can select and tune known components, but cannot produce structure outside that space. We present LACE, an AutoML framework that instead searches over complete executable pipeline programs: an evolutionary loop maintains a population of scikit-learn-compatible Python classes, and a large language model acts as the variation operator. To our knowledge, LACE is the first to formulate general tabular pipeline AutoML this way, evaluated on standardized OpenML tasks under a leakage-controlled protocol that withholds dataset identity from the generator. Because every candidate is ordinary Python, the returned pipeline and the search that produced it can be inspected and edited directly, rather than only through a framework's model objects. On 68 OpenML classification tasks, LACE with GPT-5.4-mini significantly outperforms auto-sklearn, H2O, and a fixed XGBoost baseline, with no detectable difference against AutoGluon, the strongest search-based system evaluated, while covering the full benchmark. Newer tabular foundation models are more accurate on the subset of tasks they support, but apply a fixed pretrained predictor rather than returning an editable task-specific program. LACE's contribution is therefore not raw accuracy but a search space defined by code: complete coverage, pipelines practitioners can reuse directly, and a component set extended by editing the prompt rather than the framework.

Problem

• AutoML
• search space limitation
• executable pipeline programs
• tabular data
• editability

Innovation

• LLM-driven Evolutionary Search
• Executable Pipeline Programs
• Code-based Search Space
• Editable AutoML
• Tabular Data