---

name: search-and-export description: "Configure AutoKeras search, tuners, persistence, callbacks, export_model, and saved Keras model reloads." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

AutoKeras search and export

Use this sub-skill when the request is about controlling AutoKeras search behavior, tuner selection, objectives, search directories, callbacks, persistence, overwrite, model export, or reloading saved Keras models.

Fast route

1. For smoke checks, set max_trials=1, epochs=1, overwrite=True, and an explicit scratch directory.
2. For real searches, choose a stable directory/project_name, decide whether to resume (overwrite=False) or start fresh (overwrite=True), and set an objective compatible with the metrics/losses.
3. Choose tuner from "greedy", "random", "hyperband", "bayesian", or pass a tuner class.
4. Use callbacks and epochs deliberately; epochs=None can trigger adaptive early stopping.
5. After fitting, call export_model(), save with Keras, and reload with custom_objects=ak.CUSTOM_OBJECTS when AutoKeras custom layers are present.

Read references/search-configuration.md for constructor and fit controls. Read references/tuner-reference.md for tuner names/classes and task-specific defaults. Read references/export-and-reload.md for saving and loading exported Keras models. Read references/troubleshooting.md for invalid tuner names, missing validation data, stale search directories, objective mismatches, OOM, slow searches, and custom-object reload errors.

Bundled helper

• scripts/export_tiny_model.py constructs or optionally fits/exports/reloads a tiny offline AutoKeras model. It defaults to dry-run mode; use --run-fit only when a tiny local search is acceptable.

Route elsewhere

• Image/text/structured task setup and data formats: ../task-apis/SKILL.md.
• Custom AutoModel graph topology, nodes, blocks, heads, multimodal or multitask composition: ../automodel-customization/SKILL.md.