factor_agent.config#

The campaign config: one YAML file holds every setting of a run.

data, reference and known hold quantlab components: each is a get_config() dict inline, the path of a store folder whose component.json describes it (quantlab.core.store_folder.save_component), or {store: <folder>, factor_names: [...]}, that store’s factor pinned to some of its outputs. reference lists the reference set: the factors candidates must be original against (the Zoo) and, in admission, uncorrelated with and incremental to; a factor’s factor_names selects its columns. known lists the known factors: every other existing factor a candidate is checked against for novelty (the Zoo), never one it is measured incremental to; an output also in reference is taken there only (CampaignConfig.load_factors). A malformed entry of either list is refused when the config is read; a store or class is only opened when the factors are loaded.

fields (field files, one per input family) and operators (the operator file) make the Vocabulary candidates write with; the framework has none of its own. operator_modules lists the modules, besides KunQuant and quantlab.factor, an operator may come from.

market describes the data in the prompts, mechanisms lists the mechanism types candidates belong to (by default DEFAULT_MECHANISMS), each with a one-line description, and prompts maps a template’s name (see factor_agent.prompts.TEMPLATES) to the file that replaces it.

Classes

`AcceptanceConfig`	How `Acceptance` validates, opens the hold-out and exports.
`AdmissionConfig`	What an evaluated candidate needs to be accepted.
`CampaignConfig`	Every setting of a campaign, read from one YAML file with `from_file`.
`ContextConfig`	What the prompt context holds; the run snapshot records it at the first start, and a resume may not change it.
`DataConfig`	The data the Evaluator reads.
`EvaluatorConfig`	How the Evaluator computes the round's graph.
`GateConfig`	The Gate's sandbox limits and graph constraints.
`Mechanism`	A mechanism type a candidate's hypothesis belongs to, with the line the prompts describe it by.
`NeutralizeConfig`	What ranked factors are neutralised against.
`RoleConfig`	One role's LLM: the backend name, the model and the backend's own settings.
`RoundConfig`	The size of a round, the planner's quotas and when the campaign stops.
`ScoringConfig`	How a factor's raw values become the values that are judged (`factor_agent.scoring`).
`TrialGraphConfig`	The Trial graph's parameters (`_trial_graph`), by default CosmosAlpha's (arXiv:2611.04555, Tables 6 and 7).

Functions

`load`	Rebuild a component given as a `get_config()` dict, a store folder path or a store entry.
`load_each`	Rebuild a component (see `load`), or each of a list of components.
`recorded_fertile_children`	The `context.fertile_children` a run snapshot's `settings` record; the default when they record none.