Reference

Talos exposes parameter sweeps through the Python Scan pattern and through SFD/CLI experiments. This section describes public callables, defaults, result fields and failure boundaries. Guides own end-to-end jobs; Developer owns maintenance and documentation proof.

Run a parameter sweep

Interface	Use it for
Scan	Five-argument model callbacks, candidate dictionaries, limits, result properties and persistence.
Backends	Standalone Keras, TensorFlow/tf.keras and native Torch result/serialization contracts.
Installation	Supported Python versions, framework extras and editable installation.
SFD and CLI	Explicit model/preparation/parameter functions, YAML manifests and CLI commands.

Inspect and recover trained models

Interface	Use it for
Analyze	Metric summaries, parameter tables, correlation and plots.
Predict	Explicit candidate selection, raw inference and class conversion.
Evaluate	Held-out F1 or MAE scores without retraining.
Deploy	Local ZIP packaging of a selected trained model and experiment provenance.
Restore	Trusted native and historical archive restoration.

Construct a training callback

Helper	Use it for
Generator	Repeating array batches and framework Sequence batches.
Hidden layers and shapes	Dense/Dropout layer count and width candidates.
Learning-rate normalizer	Fixed optimizer-specific scaling.
Metrics	Keras training metrics and their scientific units.
Monitoring	Round progress, parameter printing, epoch logs and training plots.
Energy draw	Endpoint GPU power samples and their watt-second estimate.
Templates	Explicit dataset acquisition, preset parameters, model callbacks and pipelines.

Explore architecture presets

Helper	Use it for
AutoParams	Generate and narrow candidate dictionaries.
AutoModel	Build a Keras training callback from architecture presets.
AutoScan	Combine presets with the Scan execution interface.
AutoPredict	Score fitted candidates and predict with the held-out winner.

Every executable fragment states its dependency and setup context. Framework extras are optional for the core, but required when a referenced operation needs that framework. The caller remains responsible for data suitability, training logic and an independent evaluation protocol.

Read next

Run the quickstart, consult Scan for an existing Talos callback, or use migration to port it to an SFD.