extern crate intel_mkl_src;

use kickscore as ks;

fn main() {
    let mut model = ks::BinaryModel::new(ks::BinaryModelObservation::Probit);

    // Spike's skill does not change over time.
    let k_spike = ks::kernel::Constant::new(0.5);

    // Tom's skill changes over time, with "jagged" (non-smooth) dynamics.
    let k_tom = ks::kernel::Exponential::new(1.0, 1.0);

    // Jerry's skill has a constant offset and smooth dynamics.
    let k_jerry: Vec<Box<dyn ks::Kernel>> = vec![
        Box::new(ks::kernel::Constant::new(1.0)),
        Box::new(ks::kernel::Matern52::new(0.5, 1.0)),
    ];

    // Now we are ready to add the items in the model.
    model.add_item("Spike", Box::new(k_spike));
    model.add_item("Tom", Box::new(k_tom));
    model.add_item("Jerry", Box::new(k_jerry));

    // At first, Jerry beats Tom a couple of times.
    model.observe(&["Jerry"], &["Tom"], 0.0);
    model.observe(&["Jerry"], &["Tom"], 0.9);

    // Then, Tom beats Spike, and then Jerry.
    model.observe(&["Tom"], &["Spike"], 1.7);
    model.observe(&["Tom"], &["Jerry"], 2.1);

    // Finally, Jerry beats Tom, and then Tom + Spike.
    model.observe(&["Jerry"], &["Tom"], 3.0);
    model.observe(&["Jerry"], &["Tom", "Spike"], 3.5);

    model.fit();

    // We can predict a future outcome...
    let (p_win, _p_los) = model.probabilities(&[&"Jerry"], &[&"Tom"], 4.0);
    println!(
        "Chances that Jerry beats Tom at t = 4.0: {:.1}%",
        100.0 * p_win
    );

    // ... or simulate what could have happened in the past.
    let (p_win, _p_los) = model.probabilities(&[&"Jerry"], &[&"Tom"], 2.0);
    println!(
        "Chances that Jerry beats Tom at t = 2.0: {:.1}%",
        100.0 * p_win
    );

    let (p_win, _p_los) = model.probabilities(&[&"Jerry"], &[&"Tom"], -1.0);
    println!(
        "Chances that Jerry beats Tom at t = -1.0: {:.1}%",
        100.0 * p_win
    );
}