ARIMA

Rust crate for ARIMA model coefficient estimation and simulation.

Please note that this crate relies on LAPACK which needs a working BLAS implementation. Per default, this crate looks for a system-installed OpenBLAS. On Debian and Ubuntu, you can install this via apt install libopenblas-base libopenblas-dev.

Example

extern crate rand;
use rand::prelude::*;
use rand::distributions::{Normal, Distribution};

use arima::{acf, sim};

fn main() {
    // initialize RNG with seed
    let mut rng: StdRng = SeedableRng::from_seed([100; 32]);

    // our noise should be normally distributed
    let normal = Normal::new(10.0, 2.0);

    // simulate time series
    let ts = sim::arima_sim(
        1000,                   // number of samples
        Some(&[0.7, 0.2]),      // AR parameters
        None,                   // MA parameters
        0,                      // difference parameter
        &|mut rng| { normal.sample(&mut rng) }, // noise fn
        &mut rng                // RNG
    ).unwrap();

    // estimate AR parameters
    let ar = acf::ar(&ts, Some(2)).unwrap();

    println!("Estimated parameters: {:?}", ar);
    // Estimated parameters: [0.7436892808499717, 0.14774749031248915]
}

Features

• Auto-correlation/covariance calculation
• Partial auto-correlation calculation
• AR parameter estimation
• Variance estimation
• ARIMA time series simulation

Roadmap

• Full ARIMA model parameter estimation
• Order estimation

License

This crate is licensed under the Apache-2.0 license.