---

name: igel description: "Use Igel for classic tabular ML, FastAPI serving, and AutoKeras-backed Auto-ML workflows." disable-model-invocation: true metadata: disco-role: operating license: MIT

Igel

Use this repo skill for the igel package: classic tabular fit/evaluate/predict workflows, serving a fitted model through FastAPI, and the AutoKeras-backed IgelCNN path.

Start here:

• references/package-overview.md for the command map, install notes, and quick route decisions.
• references/troubleshooting.md for cross-cutting legacy install/import failures.
• references/repo-provenance.md when checking whether this skill matches the current repository baseline.
• references/repo-routing-metadata.json for the router placement metadata used by managed import and selection logic.
• scripts/check_env.py for a safe import/help smoke check.

Route map

1. Classic tabular ML: create configs, fit models, evaluate, predict, run experiment, export to ONNX, inspect model/metric catalogs, or use Igel(**kwargs) from Python. Route to tabular-workflows.
2. Serving a fitted model: start the FastAPI endpoint, call /predict, or use the bundled client helper. Route to deployment.
3. AutoKeras / Auto-ML: inspect IgelCNN, supported Auto-ML task names, or image/text/structured AutoKeras workflows. Route to auto-ml.
4. Route selection is unclear: use the package overview and repository routing metadata before guessing.

Common signals

• If the user names classic sklearn-style commands such as fit, evaluate, predict, experiment, export, models, or metrics, go to tabular-workflows.
• If the user names FastAPI, /predict, host or port flags, a REST client, or model_results serving, go to deployment.
• If the user names AutoKeras, IgelCNN, ImageClassification, TextClassification, or structured-data task strings, go to auto-ml.
• If the user only wants install or import health, start with scripts/check_env.py and the cross-cutting troubleshooting reference.

Minimal install and check

The validated package version is igel 0.7.0 on Python 3.8-era dependencies. A compatible environment needs the legacy NumPy/SciPy/scikit-learn stack that this release expects; modern dependency resolvers can otherwise drift into incompatible versions.

Typical install from PyPI:

python -m pip install igel

If you are working from a local checkout, editable install is also reasonable:

python -m pip install -e .

After install, run the bundled smoke helper:

python scripts/check_env.py

For Auto-ML checks, add the helper's Auto-ML mode:

python scripts/check_env.py --auto-ml

What this skill covers

• Classic init, fit, evaluate, predict, experiment, export, models, metrics, version, and info commands.
• FastAPI serving of fitted models and the IGEL_MODEL_RESULTS_PATH contract.
• IgelCNN AutoKeras task selection and the current docs/source caveat around the missing auto-train Click command.
• Public package artifacts such as model_results/model.joblib, description.json, evaluation.json, predictions.csv, and model.onnx.

What this skill does not do

• It does not tell future agents to open the original repository docs or examples at runtime.
• It does not treat gui or the stale source Dockerfile as a verified self-contained deployment path.
• It does not expose repo-maintenance or publication workflows; those belong elsewhere.