ExperimentConfig

Top-level experiment.

Either run (one run) or runs (multiple - a campaign) must be set.

The matrix shorthand expands into runs automatically: each axis maps to a list of values, and the cartesian product is taken. Each cell inherits everything from base_run and overrides the matrixed fields.

Attributes

attributeversionint

= 1

Schema version. Bumped when breaking changes are made.

attributenamestr

attributedescriptionstr

= ''

attributeoutput_dirstr

= './outputs'

Where local artifacts (checkpoints, logs) are written.

attributedata_storeDataStoreSpec

= Field(default_factory=DataStoreSpec)

attributelog_storeLogStoreSpec

= Field(default_factory=LogStoreSpec)

attributecallbackslist[CallbackSpec]

= Field(default_factory=list)

attributerunRunConfig | None

= None

attributerunslist[RunConfig] | None

= None

attributematrixMatrixSpec | None

= None

attributecontinualContinualConfig | None

= None

attributen_repeatsint

= 1

How many seeded replicates per primary RunConfig. 1 = no grouping.

attributebase_seedint | None

= None

Starting seed for auto-generated replicates. None → use each primary's own seed.

attributeparent_experiment_idstr | None

= None

For evolutionary lineage.

attributemetadatadict[str, Any]

= Field(default_factory=dict)

Free-form (e.g. budget, hypothesis, client tag).

Functions

funcmodel_post_init(self, __context) -> None

paramself

param__contextAny

Returns

None

ExperimentConfig

Attributes

Functions

On this page