org.replikativ.foerster.amortized.compiler — org.replikativ/foerster 0.1.119

Source-derived variable schemas and attention masks for GSDM tasks. Uses Foerster's existing program analysis. Factor-scope cliques are inference connectivity, not discovered causal edges. Fixed instances only; general simulator-boundary tasks remain available for opaque/dynamic internals.

bp__org.replikativ.spindel.effects.await_SLASH_await clj

bp__org.replikativ.spindel.effects.await_SLASH_await_finalization clj

bp__org.replikativ.spindel.effects.savepoint_SLASH_savepoint clj

bp__org.replikativ.spindel.effects.track_SLASH_track clj

compile clj

(compile model)

(compile model
         {:keys [id args schema-overrides causal? targets known-inputs]
          :or {args []}
          :as options})

Compile a source-backed defn Var, analysis/model or compiler/model into a task. Derives variable names/types and moralized factor-scope attention edges. Default :conditionings :any trains arbitrary observed subsets without 2^n enumeration. :args fixes loop sizes and model inputs. :schema-overrides handles unknown types/features. :causal? explicitly declares autonomous mechanisms and enables noise reconstruction; this initially needs named sites. Unknown control flow and unexpanded loops are refused, never guessed.

model clj macro

(model args & body)

Write ordinary simulator code; this wrapper records source for analysis. Like analysis/model, the body is already inside a spin. No mask declaration.