stochastic-rs
Getting started

Workspace layout

How the stochastic-rs Cargo workspace is organised — sub-crates, dependency topology, and how the umbrella re-exports preserve v1.x import paths.

Workspace layout

stochastic-rs is a Cargo workspace with 9 sub-crates as siblings under the workspace root. The stochastic-rs umbrella crate keeps the v1.x public API by re-exporting everything via pub use.

stochastic-rs/                        (workspace root + umbrella crate)
├── stochastic-rs-core/               simd_rng (foundation)
├── stochastic-rs-distributions/      FloatExt, SimdFloatExt, 19 distributions
├── stochastic-rs-stochastic/         ProcessExt, 120+ processes
├── stochastic-rs-copulas/            BivariateExt, MultivariateExt, 10 copulas
├── stochastic-rs-stats/              MLE, Hurst, realised, stationarity, …
├── stochastic-rs-quant/              PricerExt, ModelPricer, calibration, vol surface
├── stochastic-rs-ai/                 Neural surrogates (feature-gated)
├── stochastic-rs-viz/                Plotly grid plotter
└── stochastic-rs-py/                 PyO3 cdylib (210 entries)

Dependency topology

        stochastic-rs-core (simd_rng)
                  │
                  ▼
   stochastic-rs-distributions (FloatExt, SimdFloatExt)
        ┌─────────┬─────────┐
        ▼         ▼         ▼
   stochastic   copulas   stats
        │                   │
        └────────┬──────────┘
                 ▼
            stochastic-rs-quant
              ┌───┴───┐
              ▼       ▼
             ai      viz

The PyO3 crate (stochastic-rs-py) sits on the side: it depends on the producer crates and re-exports #[pyclass]-wrapped types via the py_distribution! / py_process_*! macros plus hand-written #[pyclass] blocks for pricers and estimators.

Why the workspace split

  1. Compile time. Pulling only stochastic-rs-distributions for a distribution-sampling job avoids compiling the quant + AI surface.
  2. Feature gating. ai, openblas, cuda, metal, viz live on the crates that own the optional dependency, so a default cargo build never tries to link CUDA on a laptop with no GPU.
  3. Versioning. Sub-crates can ship patch releases independently when needed — the umbrella tracks the highest sub-crate version.

Path preservation across the workspace split

If your code imports via stochastic_rs::stochastic::diffusion::* or stochastic_rs::quant::pricing::*, no change is required when upgrading across the workspace split. The umbrella pub uses every sub-crate's public surface under the historical paths.

Quickstart

A 5-minute end-to-end tour — simulate an OU path, price a Heston call, and run a Hurst estimator. Same code shown in Rust and Python side by side.

Traits overview

The trait surface that organises the library — FloatExt, ProcessExt, DistributionExt, PricerExt, ModelPricer, Calibrator, GreeksExt, and friends.

On this page

Workspace layoutDependency topologyWhy the workspace splitPath preservation across the workspace split