---

name: task-apis description: "Use AutoKeras supervised task APIs for image, text, and structured-data classification and regression workflows." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

AutoKeras task APIs

Use this sub-skill when the user wants AutoKeras' high-level supervised task classes rather than a custom AutoModel graph:

• ak.ImageClassifier and ak.ImageRegressor
• ak.TextClassifier and ak.TextRegressor
• ak.StructuredDataClassifier and ak.StructuredDataRegressor

Fast route

1. Set the Keras backend before importing Keras or AutoKeras.
2. Pick the task class by input modality and target type.
3. Keep first runs small: max_trials=1, epochs=1, explicit directory=..., and overwrite=True for disposable smoke checks.
4. Pass data in the shape expected by the task class; use validation_data for deterministic tiny checks or a non-zero validation_split for ordinary training.
5. Call fit, then predict, evaluate, or export_model depending on the requested deliverable.

Read references/api-reference.md for verified constructor arguments and method behavior. Read references/data-formats.md when data shape, column_names, column_types, label encoding, or CSV/DataFrame handling matters. Read references/workflows.md for image, text, and structured-data recipes that avoid external downloads. Read references/troubleshooting.md when a run fails or behaves unexpectedly.

Bundled offline helpers

Run these helpers with --help first. They default to dry-run construction so future agents can inspect setup without launching a model search; add --run-fit only for a tiny local smoke run.

• scripts/run_tiny_image_task.py constructs or runs an image classifier/regressor on synthetic arrays.
• scripts/run_tiny_text_task.py constructs or runs a text classifier/regressor on synthetic full-sentence strings.
• scripts/run_tiny_structured_task.py constructs or runs a structured-data classifier/regressor on tiny mixed tabular data.

Route elsewhere

• Custom topologies, blocks, multimodal inputs, multitask outputs, or manual AutoModel graphs: ../automodel-customization/SKILL.md.
• Tuner selection, search persistence, overwrite, objectives, callbacks, export_model, or reloading saved Keras models: ../search-and-export/SKILL.md.
• Benchmark-scale timing, downloaded public datasets, Colab/Drive workflows, Docker, or release automation are not part of this operating sub-skill.