reservoir-train

Crates.ioreservoir-train
lib.rsreservoir-train
version0.1.0
created_at2025-12-22 16:26:53.818511+00
updated_at2025-12-22 16:26:53.818511+00
descriptionTraining utilities for Echo State Networks: ridge/lasso readout solvers, ESN builder helpers, and optional static code generation for embedded inference.
homepagehttps://github.com/haradama/reservoir-rs/tree/main/crates/train
repositoryhttps://github.com/haradama/reservoir-rs/tree/main/crates/train
max_upload_size
id1999984
size66,535
haradama (haradama)

documentation

https://docs.rs/reservoir-train

README

reservoir-rs

A minimal, experimental Rust workspace for reservoir computing / Echo State Networks (ESNs).

This project focuses on:

  • reusable core traits (reservoir / readout / trainer)
  • optional no_std support via feature flags
  • simple datasets for time-series experiments
  • lightweight training + inference components

Note: This is early-stage software. APIs, module layout, and crate names may change.

What’s inside (high level)

  • Core: traits, scalar/type aliases, basic metrics (MSE/RMSE/NRMSE/R²)
  • Inference: ESN-style reservoir + readout building blocks (dynamic and/or const-generic variants depending on features)
  • Training: simple readout training (e.g., ridge / lasso-style solvers)
  • Datasets: synthetic time-series generators (for quick experiments)
  • Embedded/QEMU tests: optional no_std inference checks on emulated targets

Build

This is a Cargo workspace.

cargo build
cargo test

Examples (if available in the workspace):

cargo run --example <name>

no_std / features

Some components can be built without the standard library.

  • Use default-features = false when you want no_std.
  • Enable alloc when you need heap-backed vectors/matrices.
  • Enable libm when targeting platforms without std floating-point support.

Exact feature sets may change as the project evolves.

License

Apache License 2.0

Commit count: 0

cargo fmt