factor_agent.config.ContextConfig

class ContextConfig(recent_rounds=5, feedback=True, theme_correlation=None, send_back_repeats=False, lineage=False, fertile_children=3)

Bases: object

What the prompt context holds; the run snapshot records it at the first start, and a resume may not change it.

Attributes

ContextConfig.feedback = True

Whether the planner and the proposer read the earlier trials (the memory, the history, the last round and the accepted factors) and a memory is distilled. Off, each round is planned and written from the data and the seed directions alone, every direction new: the control of a test of whether the search learns from its trials.

ContextConfig.fertile_children = 3

The best child outcomes a Lineage’s fertility is the mean of, in the prompts and the Trial graph alike.

ContextConfig.lineage = False

Whether the planner’s accepted factors and the proposer’s parent block carry each trial’s Lineage (_lineage): how the trials built on it did; it needs feedback.

ContextConfig.recent_rounds = 5

Rounds whose trials the planner reads one line each (the recent history); older trials are counted by mechanism type and outcome (the earlier summary).

ContextConfig.send_back_repeats = False

Whether a new direction’s candidate as correlated with an earlier trial is sent back to be rewritten once before it is evaluated; it needs theme_correlation.

ContextConfig.theme_correlation = None

The correlation, in absolute value, at which an evaluated trial is grouped with an earlier one (_feedback.theme_groups), the groups the planner reads; None (the default) groups none.