RunContext

Everything an algorithm needs to execute one training run.

Held loosely on purpose: an algorithm only consumes the fields it needs. Backends construct this from the parsed ExperimentConfig.

Attributes

attributerun_idstr

Stable id for this run; used for output paths and log keys.

attributeoutput_dirstr

Local filesystem dir where the run can write checkpoints, logs, etc.

attributeconfigAny

The full parsed ExperimentConfig (kept generic to avoid circular import).

attributedata_storeDataStore

Datastore handle (read inputs, write outputs).

attributelog_storeLogStore

Log store (metrics, scalars, artifacts).

attributebackendBackend

Backend handle (Tinker / Local / Mock).

attributeextrasdict[str, Any]

= field(default_factory=dict)

Free-form bag for backend-specific bits (e.g. tinker training_client).

func__init__(self, run_id, output_dir, config, data_store, log_store, backend, extras=dict()) -> None

paramself

paramrun_idstr

paramoutput_dirstr

paramconfigAny

paramdata_storeDataStore

paramlog_storeLogStore

parambackendBackend

paramextrasdict[str, Any]

= dict()

Returns

None