1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146

#![allow(dead_code)]
#[allow(unused_imports)]
use itertools::Itertools;

use crate::calculus::differentiation::*;
use crate::linalg::vector::Vector;
pub type Function<'a> = &'a dyn Fn(&[f64], &[f64]) -> f64;
pub type Cost<'a> = &'a dyn Fn(&dyn Fn(&[f64], &[f64]) -> f64, &[f64], &[Vec<f64>], &[f64]) -> f64;
pub type CostWeighted<'a> =
    &'a Fn(&dyn Fn(&[f64], &[f64]) -> f64, &[f64], &[Vec<f64>], &[f64], &[f64]) -> f64;

pub fn least_squares(f: Function, params: &[f64], vars: &[Vec<f64>], y: &[f64]) -> f64 {
    #[allow(clippy::cast_precision_loss)]
    vars.iter()
        .zip(y.iter())
        .map(|(x, y)| (y - f(x, &params)).powf(2.0))
        .map(|x| x / vars.len() as f64)
        .sum()
}
pub fn weighted_least_squares(
    f: Function,
    params: &[f64],
    vars: &[Vec<f64>],
    y: &[f64],
    weights: &[f64],
) -> f64 {
    #[allow(clippy::cast_precision_loss)]
    vars.iter()
        .zip(y.iter())
        .zip(weights.iter())
        .map(|((x, y), w)| w * (y - f(x, &params).powf(2.0)))
        .map(|x| x / vars.len() as f64)
        .sum()
}

#[allow(clippy::too_many_arguments)]
pub fn fit_weighted(
    f: Function,
    vars: &[Vec<f64>],
    y: &[f64],
    weights: &[f64],
    bounds: (Vec<f64>, Vec<f64>),
    cost_function: Option<CostWeighted>,
    threshold: Option<f64>,
    learning_rates: (Option<f64>, Option<f64>),
    iterations: Option<usize>,
) -> (Vec<f64>, f64) {
    let cost_function = cost_function.unwrap_or(&weighted_least_squares);
    let cost_func: Cost = &(|f, params, vars, y| cost_function(f, params, vars, y, weights));
    fit(
        f,
        vars,
        y,
        bounds,
        Some(&cost_func),
        threshold,
        learning_rates,
        iterations,
    )
}

#[allow(clippy::too_many_arguments)]
pub fn fit(
    f: Function,
    vars: &[Vec<f64>],
    y: &[f64],
    bounds: (Vec<f64>, Vec<f64>),
    cost_function: Option<Cost>,
    threshold: Option<f64>,
    learning_rates: (Option<f64>, Option<f64>),
    iterations: Option<usize>,
) -> (Vec<f64>, f64) {
    let cost_function = cost_function.unwrap_or(&least_squares);
    let threshold = threshold.unwrap_or(0.5);
    let eta = learning_rates.0.unwrap_or(0.1);
    let gamma = learning_rates.1.unwrap_or(0.9);
    let iterations = iterations.unwrap_or(100_000);
    let cost_function_params = |params: &Vec<f64>| cost_function(f, params, vars, y);
    //Use diff to minimise cost
    let mut params = bounds.0;

    let mut _found = false;
    let mut delta = Vector {
        data: vec![0.0; params.len()],
    };
    for _ in 0..iterations {
        let gradient = diff(
            &cost_function_params,
            &(Vector::from(&params) - gamma * delta.clone()).data,
            0.001,
        )
        .unit();
        let cost = cost_function_params(&params);
        if cost < threshold {
            _found = true;
            break;
        }
        delta = gamma * delta + eta * gradient.clone();
        params = (Vector::from(&params) - delta.clone()).data;
    }
    let cost = cost_function_params(&params);
    (params, cost)
}

#[cfg(test)]
mod tests {
    #[allow(unused_imports)]
    use super::*;
    #[test]
    fn test_quadratic_fit() {
        let vars: Vec<Vec<f64>> = (0..20).map(f64::from).map(|x| vec![x]).collect();
        let y: Vec<f64> = (0..20)
            .map(f64::from)
            .map(|x| 5.0 * x.powf(2.0) + 2.0 * x + -8.0)
            .collect();
        let bounds = (vec![-10.0, -10.0, -10.0], vec![100.0, 100.0, 100.0]);
        let f = &(|vars: &[f64], params: &[f64]| {
            params[0] * vars[0].powf(2.0) + params[1] * vars[0] + params[2]
        });
        println!(
            "{:?}",
            fit(f, &vars, &y, bounds, None, None, (None, None), None)
        );
    }
    #[test]
    fn test_2d_fit() {
        let vars: Vec<Vec<f64>> = (0..20)
            .cartesian_product(0..20)
            .map(|(x, y)| vec![f64::from(x), f64::from(y)])
            .collect();
        let y: Vec<f64> = (0..20)
            .cartesian_product(0..20)
            .cartesian_product(0..20)
            .map(|((x, y), z)| (f64::from(x), f64::from(y), f64::from(z)))
            .map(|(x, y, z)| 2.0 * x.powf(2.0) + 3.0 * y.powf(2.0) + 4.0 * z.powf(2.0))
            .collect();
        let bounds = (vec![-10.0, -10.0, -10.0], vec![50.0, 50.0, 50.0]);
        let f = &(|vars: &[f64], params: &[f64]| {
            params[0] * vars[0].powf(2.0) + params[1] * vars[1].powf(2.0)
        });
        println!(
            "{:?}",
            fit(f, &vars, &y, bounds, None, None, (None, None), None)
        );
    }
}