miniboosts 0.2.1

MiniBoosts: A collection of boosting algorithms written in Rust 🦀
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

use grb::prelude::*;


use crate::Sample;
use crate::hypothesis::Classifier;

/// A linear programming model for edge minimization. 
pub(super) struct LPModel {
    pub(self) model: Model,
    pub(self) gamma: Var,
    pub(self) dist: Vec<Var>,
    pub(self) constrs: Vec<Constr>,
}


impl LPModel {
    /// Initialize the LP model.
    /// arguments.
    /// - `size`: Number of variables (Number of examples).
    /// - `upper_bound`: Capping parameter. `[1, size]`.
    pub(super) fn init(size: usize, upper_bound: f64) -> Self {
        let mut env = Env::new("").unwrap();
        env.set(param::OutputFlag, 0).unwrap();

        let mut model = Model::with_env("MLPBoost", env).unwrap();


        // Set GRBVars
        let gamma = add_ctsvar!(model, name: "gamma", bounds: ..)
            .unwrap();

        let dist = (0..size).map(|i| {
                let name = format!("d[{i}]");
                add_ctsvar!(model, name: &name, bounds: 0.0..upper_bound)
                    .unwrap()
            }).collect::<Vec<Var>>();


        // Set a constraint
        model.add_constr(&"sum_is_1", c!(dist.iter().grb_sum() == 1.0))
            .unwrap();


        // Set objective function
        model.set_objective(gamma, Minimize).unwrap();


        // Update the model
        model.update().unwrap();


        Self {
            model,
            gamma,
            dist,
            constrs: Vec::new(),
        }
    }


    /// Solve the edge minimization problem 
    /// over the hypotheses `h1, ..., ht` 
    /// and outputs the optimal value.
    pub(super) fn update<F>(
        &mut self,
        sample: &Sample,
        clf: &F
    ) -> f64
        where F: Classifier
    {
        // If we got a new hypothesis,
        // 1. append a constraint, and
        // 2. optimize the model.
        let edge = sample.target()
            .into_iter()
            .enumerate()
            .map(|(i, y)| y * clf.confidence(sample, i))
            .zip(self.dist.iter().copied())
            .map(|(yh, d)| d * yh)
            .grb_sum();


        let name = format!("{t}-th hypothesis", t = self.constrs.len());


        self.constrs.push(
            self.model.add_constr(&name, c!(edge <= self.gamma))
                .unwrap()
        );


        self.model.update()
            .unwrap();


        self.model.optimize()
            .unwrap();


        let status = self.model.status().unwrap();
        if status != Status::Optimal {
            panic!("Status is {status:?}. Something wrong.");
        }


        self.model.get_obj_attr(attr::X, &self.gamma)
            .unwrap()
    }

    /// Returns the distribution over examples.
    pub(super) fn distribution(&self)
        -> Vec<f64>
    {
        self.dist.iter()
            .map(|d| self.model.get_obj_attr(attr::X, d).unwrap())
            .collect::<Vec<_>>()
    }


    /// Returns the weights over the hypotheses.
    pub(super) fn weight(&self) -> impl Iterator<Item=f64> + '_
    {
        self.constrs[0..].iter()
            .map(|c| self.model.get_obj_attr(attr::Pi, c).unwrap().abs())
    }
}