Skip to main content

limen.sfd.rule_based

Configuration schema and predicate compiler for non-learned, rule-based decoders.

Canonical docs​

What this package owns​

Owns the rule-based decoder building blocks: the RuleBasedConfig schema (with leaf-condition validation) and build_predicate, which compiles a condition tree into an evaluable predicate over a frame. Does not own the SFD wiring that uses these (owned by the rule_based modules in limen.sfd.foundational_sfd / reference_architecture) or experiment execution.

Key entry points​

Entry pointUse caseNotes
RuleBasedConfigValidated rule-strategy configurationEnforces required fields per leaf condition type
build_predicateCompile a condition tree into a callable predicateEvaluated against a pl.DataFrame

Adjacent modules​

  • limen.sfd.reference_architecture.rule_based (RuleBasedStrategy) and limen.sfd.foundational_sfd.rule_based build on these.
  • limen.data.utils.split_data_to_rule_based_prep_output produces the data_dict variant that keeps the raw columns predicates reference.

Quick orientation​

rule_based/
├── config.py       # RuleBasedConfig + leaf-condition validation
└── predicates.py   # build_predicate + operator table

Things to know​

  • Leaf conditions are validated by type: each of threshold, relative, crossover, slope, and sql_expr has its own required fields, and a missing field raises ValueError.
  • Logical nodes combine leaves with and, or, and not.
  • A condition may specify persistence_n or recency_n, but not both.
  • Predicate values support {param}-style formatting, so thresholds can be parameterized from the manifest.