specparam-rs

A Rust implementation of the specparam (fooof). Procedures are identical to specparam in Python, with the exception that argmin is used for optimization, rather than scipy's curve_fit. Performance tests show that this implementation is 30-50x faster than the python implementation, depending on simulation parameters.

Quickstart

Install Rust

brew install rust

Python bindings via PyO3:

git clone [email protected]:ryanhammonds/specparam-rs.git
cd specparam-rs

pip install maturin

maturin develop --release

from specparam import SpecParam
sp = SpecParam(max_n_peaks=5, aperiodic_mode='linear')
res = sp.fit(freqs, powers)

Or, using pure Rust:

use ndarray::{Array1};

mod gen;
mod optimization;
mod specparam;

use gen::{lorentzian, linear, peak, noise};
use specparam::SpecParam;

fn main() {
    // Frequencies
    let freqs : Array1<f64> = Array1::range(1., 101., 1.0);

    // Powers : lorentzian + oscillation + noise
    let mut powers : Array1<f64> = lorentzian(&freqs, 10.0, 2.0, 100.0);
    let sig_noise : Array1<f64> = noise(&freqs, 0.05);

    powers = (powers + sig_noise).mapv(|p| (10.0_f64).powf(p));
    powers = powers + peak(&freqs, 20.0, 20.0, 2.0);

    // Initialize & fit
    let mut sp = SpecParam{
        max_n_peaks: 1,
        aperiodic_mode: "lorentzian".to_string(),
        ..Default::default()
    };

    let results = sp.fit(&freqs, &powers);
}

And then to excute from a shell:

cargo run --release