fanova 0.3.0

A Rust implementation of fANOVA (functional analysis of variance)
Documentation 
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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165

use crate::decision_tree::{DecisionTreeOptions, DecisionTreeRegressor};
#[cfg(test)]
use crate::functions;
use crate::table::Table;
use rand::rngs::StdRng;
use rand::{Rng, RngExt, SeedableRng};
use rayon::iter::{IntoParallelIterator, ParallelIterator};
use std::num::NonZeroUsize;

/// Random forest options.
#[derive(Debug, Clone)]
pub struct RandomForestOptions {
    trees: NonZeroUsize,
    max_features: Option<NonZeroUsize>,
    seed: Option<u64>,
}

impl RandomForestOptions {
    /// Makes a `RandomForestOptions` instance with the default settings.
    pub fn new() -> Self {
        Self::default()
    }

    /// Sets the random generator seed.
    ///
    /// The default value is random.
    pub fn seed(mut self, seed: u64) -> Self {
        self.seed = Some(seed);
        self
    }

    /// Sets the number of decision trees.
    ///
    /// The default value is `100`.
    pub fn trees(mut self, trees: NonZeroUsize) -> Self {
        self.trees = trees;
        self
    }

    /// Sets the number of maximum candidate features used to determine each decision tree node.
    ///
    /// The default value is `sqrt(the number of features)`.
    pub fn max_features(mut self, max: NonZeroUsize) -> Self {
        self.max_features = Some(max);
        self
    }
}

impl Default for RandomForestOptions {
    fn default() -> Self {
        Self {
            trees: NonZeroUsize::new(100).expect("never fails"),
            max_features: None,
            seed: None,
        }
    }
}

impl RandomForestOptions {
    fn tree_rngs(&self) -> impl Iterator<Item = StdRng> {
        let seed = self.seed.unwrap_or_else(|| rand::rng().random());
        let mut rng = StdRng::seed_from_u64(seed);
        (0..self.trees.get()).map(move |_| StdRng::seed_from_u64(rng.random()))
    }
}

#[derive(Debug)]
pub struct RandomForestRegressor {
    forest: Vec<DecisionTreeRegressor>,
}

impl RandomForestRegressor {
    pub fn fit(table: Table, options: RandomForestOptions) -> Self {
        let max_features = Self::decide_max_features(&table, &options);
        let forest = options
            .tree_rngs()
            .map(|mut rng| Self::tree_fit(&mut rng, &table, max_features))
            .collect::<Vec<_>>();
        Self { forest }
    }

    pub fn fit_parallel(table: Table, options: RandomForestOptions) -> Self {
        let max_features = Self::decide_max_features(&table, &options);
        let forest = options
            .tree_rngs()
            .collect::<Vec<_>>()
            .into_par_iter()
            .map(|mut rng| Self::tree_fit(&mut rng, &table, max_features))
            .collect::<Vec<_>>();
        Self { forest }
    }

    fn decide_max_features(table: &Table, options: &RandomForestOptions) -> usize {
        if let Some(n) = options.max_features {
            n.get()
        } else {
            (table.features_len() as f64).sqrt().ceil() as usize
        }
    }

    fn tree_fit<R: Rng + ?Sized>(
        rng: &mut R,
        table: &Table,
        max_features: usize,
    ) -> DecisionTreeRegressor {
        let table = table.bootstrap_sample(rng);
        let tree_options = DecisionTreeOptions {
            max_features: Some(max_features),
        };
        DecisionTreeRegressor::fit(rng, table, tree_options)
    }

    pub fn into_trees(self) -> Vec<DecisionTreeRegressor> {
        self.forest
    }

    #[cfg(test)]
    fn predict(&self, xs: &[f64]) -> f64 {
        functions::mean(self.forest.iter().map(|tree| tree.predict(xs)))
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn regression_works() -> Result<(), Box<dyn std::error::Error>> {
        let columns = [
            // Features.
            &[
                0.0, 0.0, 1.0, 2.0, 2.0, 2.0, 1.0, 0.0, 0.0, 2.0, 0.0, 1.0, 1.0, 2.0,
            ],
            &[
                2.0, 2.0, 2.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 2.0, 1.0,
            ],
            &[
                1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0,
            ],
            &[
                0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0,
            ],
            // Target.
            &[
                25.0, 30.0, 46.0, 45.0, 52.0, 23.0, 43.0, 35.0, 38.0, 46.0, 48.0, 52.0, 44.0, 30.0,
            ],
        ];
        let train_len = columns[0].len() - 2;

        let table = Table::new(columns.iter().map(|f| &f[..train_len]).collect())?;

        let options = RandomForestOptions::default().seed(0);
        let regressor = RandomForestRegressor::fit(table, options);
        assert_eq!(
            regressor.predict(&columns.iter().map(|f| f[train_len]).collect::<Vec<_>>()),
            41.221166666666676
        );
        assert_eq!(
            regressor.predict(&columns.iter().map(|f| f[train_len + 1]).collect::<Vec<_>>()),
            44.51380952380952
        );

        Ok(())
    }
}