Configuration

Every setting of the pipeline and the analysis is a field of Config; the profiles are class methods. How to change them: Configuration.

marathonts.bench.config.Config dataclass

profile classmethod

profile(name: str = 'min', **kw) -> 'Config'

min (Colab run), mid (min + the VAE per seed) or max (everything on); see the module docstring.

quick classmethod

quick(**kw) -> 'Config'

params

params(name: str) -> dict

Parameters of the component name (extra[name]), a copy; {} if none are set.

to_json

to_json() -> str

Every field as JSON (paths as strings): the file --config reads and collect saves with the tables.

from_json classmethod

from_json(text: str) -> 'Config'

Inverse of to_json (used to hand the whole config to another environment).

unit_fingerprint

unit_fingerprint(*components: str, upstream=(), files=(), dataset: str | None = None) -> str

Hash of what one unit of work depends on: the processed data, the settings of the given components, fingerprints of the units it reads (upstream) and the contents of the exogenous files ("targets", "models"). Stored in the unit's record; a mismatch means "redo". With dataset, the thresholds file counts with the built-in datasets and that dataset only: the thresholds an added dataset brings do not redo the units of the others.

fingerprint

fingerprint() -> str

Short hash of the settings that determine the artifacts (scope and paths excluded).