aideml

Why we included this project

AIDE ML is for anyone who regularly needs to turn a raw dataset into a working model without hand-writing every experiment. Point it at the data, describe the goal and the metric you care about in plain English, and it iterates on candidate code, fixing failures and benchmarking results until the number you chose improves. It's built as a lean, research-oriented package, so agent-architecture researchers can swap in their own search heuristics or evaluators, while practitioners get a fast path from data to a competitive pipeline without writing grid-search glue. The algorithm is described in a published paper, making this a solid reference implementation for understanding how a coding agent can systematically search a solution space. Teams prototyping AutoML-style workflows or studying agent behavior on code tasks will find it a compact, inspectable starting point.