Expand description
Next-app prediction benchmark for DiPECS.
Provides a reproducible backtest harness that evaluates DiPECS decision backends and simple baselines against ground-truth labels derived from synthetic Android traces.
Re-exports§
pub use runner::run_benchmark;pub use runner::BenchmarkRunConfig;pub use types::BackendMetrics;pub use types::BenchmarkConfig;pub use types::BenchmarkReport;pub use types::ExclusionCounts;pub use types::LatencySummary;pub use types::NextAppLabel;pub use types::NextAppPredictor;pub use types::PredictionResult;pub use types::ScoredPrediction;
Modules§
- baselines
- Simple baseline predictors for the next-app benchmark.
- context_
loader - Load labels and build per-label
StructuredContextwindows from raw traces. - metrics
- Metric aggregation for the next-app benchmark.
- predictor
- Adapt DiPECS decision backends into the
NextAppPredictortrait. - report
- Report serialization helpers.
- runner
- Benchmark runner: train/test split, per-scenario evaluation, and report generation.
- types
- Types for the next-app prediction benchmark.