akanalytics / spsa

Simultaneous perturbation stochastic approximation implemented in Rust

spsa

Simultaneous perturbation stochastic approximation implemented in Rust.

• Fast convergence
• Black-box, derivative-free optimization
• Stochastic functions (can be used on noisy functions)
• Converges well in higher-dimensions
• Hard constraints by returning NaN
• Automatic learning rate tuning and adaptive moment estimation
• Reusable allocation (does not allocate during optimization process)
• SIMD optimization

Based on this implementation by SimpleArt

API reference (docs.rs)

Usage

Add this to your Cargo.toml:

[dependencies]
spsa = "0.2.1"

License

spsa is distributed under the terms of both the MIT license and the Apache License (Version 2.0).

See LICENSE-APACHE and LICENSE-MIT for details.