Expand description
Execution engine for Soma computational graphs.
Provides the runtime components that execute compiled plans:
runner— trait-based execution: LocalRunner, StreamExecutor, StudyRunner, PbtRunnerexecutor— walks [ExecutionPlan] trees (sequence, parallel, cached, remote)GraphSession— the primary orchestrator: Graph + filters → compile → executecache— LRU memory cache, local disk cache, tiered cachesampler— hyperparameter samplers (Grid, Random, Bayesian/TPE)pruner— early stopping strategies (Median, Percentile)
Re-exports§
pub use cache::LocalCache;pub use cache::MemoryCache;pub use cache::TieredCache;pub use event_bus::EventBus;pub use executor::Context;pub use executor::GraphInfo;pub use executor::execute;pub use executor::resolve_input;pub use executors::FittedFilter;pub use executors::FnPbtExecutor;pub use executors::FnTrialExecutor;pub use executors::PbtConfig;pub use executors::PbtExecutor;pub use executors::PbtRunner;pub use executors::PopulationMember;pub use executors::StreamExecutor;pub use executors::StudyRunner;pub use executors::TrialExecutor;pub use executors::TrialOutcome;pub use filter_library::FilterLibrary;pub use forward::Batched;pub use forward::ForwardStrategy;pub use forward::Standard;pub use forward::Stream;pub use graph_session::GraphSession;pub use graph_session::graph_fit;pub use graph_session::graph_predict;pub use graph_session::graph_run;pub use pruner::MedianPruner;pub use pruner::PercentilePruner;pub use pruner::Pruner;pub use runner::LocalRunner;pub use runner::RemoteRunner;pub use runner::Runner;pub use runner::Transport;pub use sampler::BayesianSampler;pub use sampler::GridSampler;pub use sampler::RandomSampler;pub use sampler::Sampler;
Modules§
- cache
- Cache implementations for the Soma runtime.
- event_
bus - Broadcast event bus for runtime observability.
- executor
- Plan executor — walks
ExecutionPlantrees and runs filter nodes. - executors
- Executors — high-level execution loops that USE a Runner.
- filter_
library - Unified filter registry — holds implementations, metadata, and trained states.
- forward
- Forward execution strategies for [
GraphSession]. - graph_
session - Graph session — the primary orchestrator for Graph → Compile → Execute.
- pruner
- Early stopping strategies for optimization studies.
- runner
- Runner module — trait-based execution contracts.
- sampler
- Hyperparameter samplers for optimization studies.