Skip to main content

limen.experiment.reducer

Feedback reducers that prune, trim, or refocus the parameter domain as an experiment runs.

Canonical docs​

What this package owns​

Owns the reducer strategies that consume in-flight experiment results and mutate the search — trimming to budget, pruning correlated or saturated regions, and refocusing on promising areas — plus the registry and shared filter/action constants. Does not own the parameter domain or the search strategies it steers.

Key entry points​

Entry pointUse caseNotes
REDUCER_REGISTRYName-to-reducer lookupResolves reducers named in a manifest
BudgetReducerCap the remaining permutation budgetTrim via TRIM_RANDOM or TRIM_WORST_FIRST
CorrelationReducerRemove wrong-direction numeric values or suggest low-impact keepsUses bootstrap correlation and sign stability
FocusReducerRefocus the domain on promising regionsNarrows around strong rounds
SaturationReducerDown-sample saturated valuesUses reversible named filters
SanityReducerGuard against degenerate configurationsFilters obviously bad combinations
PruningStrategyBase pruning behaviorEmits ACTION_SUGGEST-style actions

Adjacent modules​

  • limen.experiment.param_domain is the domain reducers mutate.
  • limen.experiment.param_search strategies observe those mutations and change what they yield next.
  • limen.experiment (UEL) invokes reducers between rounds with accumulated results.

Quick orientation​

reducer/
├── budget_reducer.py       # BudgetReducer (+ TRIM_* constants)
├── correlation_reducer.py  # CorrelationReducer
├── focus_reducer.py        # FocusReducer
├── saturation_reducer.py   # SaturationReducer
├── sanity_reducer.py       # SanityReducer
├── pruning_strategy.py     # PruningStrategy (+ ACTION_SUGGEST)
├── filter_types.py         # FILTER_* constants
└── registry.py             # REDUCER_REGISTRY

Things to know​

  • Reducers inspect the log and MSQ but return declarative interventions without mutating either. FeedbackController dispatches those interventions to MSQ, then notifies the active search strategy.
  • filter_types.py defines the FILTER_* constants (FILTER_KEEP_VALUES, FILTER_KEEP_BETWEEN, FILTER_EXCLUDE_VALUE, FILTER_SAMPLE) shared across reducers.
  • Reducers are opt-in through the manifest and only run on the advanced search path.