---

name: auto-ml description: "Use Igel's AutoKeras-backed Auto-ML helpers for image, text, and structured-data task selection, IgelCNN train/evaluate/predict behavior, and current Auto-ML CLI caveats." metadata: disco-role: operating disable-model-invocation: true license: MIT

Auto-ML

Use this sub-skill when the task is about Igel's AutoKeras path: IgelCNN, igel.auto.models.Models.get, image classification/regression, text classification/regression, structured-data AutoKeras task names, or the documented-but-missing igel auto-train command.

Start here:

• Use references/api-reference.md for exact IgelCNN constructor/method behavior, save/load assumptions, and supported task names.
• Use references/workflows.md for the safe operating sequence, directory layouts, and bounded examples.
• Use references/troubleshooting.md when imports, task names, data layout, artifacts, or CLI expectations fail.
• Use scripts/inspect_auto_ml.py for an import/signature/task-selector smoke check that does not train, download data, or require the original source checkout.

Route boundaries:

1. Route classic CSV/tabular igel fit, evaluate, predict, experiment, models, metrics, and ONNX/export work to tabular-workflows; those are not AutoKeras workflows.
2. Route repo-wide skill selection back to the root router when the user is not clearly asking for Auto-ML.
3. Route FastAPI serving, Docker, GUI, and client-request tasks away from this sub-skill; this sub-skill only covers the AutoKeras-backed training/evaluation/prediction helpers and task registry.
4. Treat AutoKeras training as potentially expensive. Prefer import/signature and task-selector checks before starting any run, and never make dataset downloads or long tuning loops the default helper path.
5. Do not rely on igel auto-train as a real current Click CLI command. The docs mention it, but the current CLI surface does not expose that command; use the programmatic IgelCNN path or route to classic tabular commands when appropriate.