jaime 2.3.1

j.a.i.m.e. is an ergonomic all purpose gradient descent engine
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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215

use std::array;
use std::time::Instant;

use crate::dual::Dual;
use crate::simd_arr::dense_simd::DenseSimd;
use crate::simd_arr::hybrid_simd::{CriticalityCue, HybridSimd};
use crate::simd_arr::SimdArr;
use rand::{Rng, SeedableRng};
use rand_chacha::ChaCha8Rng;

use super::Minimizer;

/// The AdamTrainer struct manages the training lifecycle using the Adam optimization algorithm.
#[derive(Clone)]
pub struct AdamMinimizer<
    const P: usize,
    ExtraData: Sync + Clone,
    S: SimdArr<P>,
    FG: Fn(&[Dual<P, S>; P], &ExtraData) -> Dual<P, S> + Sync + Clone,
    F: Fn(&[f32; P], &ExtraData) -> f32 + Sync + Clone,
    ParamTranslate: Fn(&[f32; P], &[f32; P]) -> [f32; P] + Clone,
> {
    cost_gradient: FG,
    cost: F,
    params: [Dual<P, S>; P],
    m: [f32; P], // First moment vector
    v: [f32; P], // Second moment vector
    t: usize,    // Time step
    param_translator: ParamTranslate,
    extra_data: ExtraData,
    last_cost: Option<f32>,
    cost_stagnation_value: Option<f32>,
    cost_stagnation_time: usize,
    cost_stagnation_threshold: f32,
    max_cost_stagnation_time: usize,
}

impl<
        const P: usize,
        ExtraData: Sync + Clone,
        FG: Fn(&[Dual<P, DenseSimd<P>>; P], &ExtraData) -> Dual<P, DenseSimd<P>> + Sync + Clone,
        F: Fn(&[f32; P], &ExtraData) -> f32 + Sync + Clone,
        ParamTranslate: Fn(&[f32; P], &[f32; P]) -> [f32; P] + Clone,
    > AdamMinimizer<P, ExtraData, DenseSimd<P>, FG, F, ParamTranslate>
{
    pub fn new_dense(
        trainable: F,
        trainable_gradient: FG,
        param_translator: ParamTranslate,
        extra_data: ExtraData,
        cost_stagnation_threshold: f32,
        max_cost_stagnation_time: usize,
    ) -> Self {
        let mut rng = ChaCha8Rng::seed_from_u64(2);

        Self {
            cost_gradient: trainable_gradient,
            cost: trainable,
            params: array::from_fn(|i| Dual::new_param(rng.gen::<f32>() - 0.5, i)),
            m: [0.0; P],
            v: [0.0; P],
            t: 0,
            param_translator,
            extra_data,
            last_cost: None,
            cost_stagnation_value: None,
            cost_stagnation_threshold,
            cost_stagnation_time: 0,
            max_cost_stagnation_time,
        }
    }
}

impl<
        const P: usize,
        const CRITICALITY: usize,
        ExtraData: Sync + Clone,
        FG: Fn(
                &[Dual<P, HybridSimd<P, CRITICALITY>>; P],
                &ExtraData,
            ) -> Dual<P, HybridSimd<P, CRITICALITY>>
            + Sync
            + Clone,
        F: Fn(&[f32; P], &ExtraData) -> f32 + Sync + Clone,
        ParamTranslate: Fn(&[f32; P], &[f32; P]) -> [f32; P] + Clone,
    > AdamMinimizer<P, ExtraData, HybridSimd<P, CRITICALITY>, FG, F, ParamTranslate>
{
    pub fn new_hybrid(
        _: CriticalityCue<CRITICALITY>,
        trainable: F,
        trainable_gradient: FG,
        param_translator: ParamTranslate,
        extra_data: ExtraData,
        cost_stagnation_threshold: f32,
        max_cost_stagnation_time: usize,
    ) -> Self {
        let mut rng = ChaCha8Rng::seed_from_u64(2);

        Self {
            cost_gradient: trainable_gradient,
            cost: trainable,
            params: array::from_fn(|i| Dual::new_param(rng.gen::<f32>() - 0.5, i)),
            m: [0.0; P],
            v: [0.0; P],
            t: 0,
            param_translator,
            extra_data,
            last_cost: None,
            cost_stagnation_value: None,
            cost_stagnation_threshold,
            cost_stagnation_time: 0,
            max_cost_stagnation_time,
        }
    }
}

impl<
        const P: usize,
        ExtraData: Sync + Clone,
        S: SimdArr<P>,
        FG: Fn(&[Dual<P, S>; P], &ExtraData) -> Dual<P, S> + Sync + Clone,
        F: Fn(&[f32; P], &ExtraData) -> f32 + Sync + Clone,
        ParamTranslate: Fn(&[f32; P], &[f32; P]) -> [f32; P] + Clone,
    > Minimizer<P> for AdamMinimizer<P, ExtraData, S, FG, F, ParamTranslate>
{
    fn get_last_cost(&self) -> Option<f32> {
        self.last_cost
    }

    fn train_step<const VERBOSE: bool>(&mut self, learning_rate: f32) {
        let t0 = Instant::now();

        let cost = (self.cost_gradient)(&self.params, &self.extra_data);

        let raw_gradient = cost.get_gradient();
        let gradient_size = f32::max(
            raw_gradient.iter().fold(0., |acc, elm| acc + (elm * elm)),
            1e-30,
        );

        // Update the time step
        self.t += 1;

        // Update the first and second moment estimates
        for i in 0..P {
            self.m[i] = 0.9 * self.m[i] + 0.1 * raw_gradient[i]; // beta1 = 0.9, beta2 = 0.999
            self.v[i] = 0.999 * self.v[i] + 0.001 * raw_gradient[i].powi(2); // epsilon = 1e-8
        }

        // Compute bias-corrected first and second moment estimates
        let m_hat: [f32; P] = array::from_fn(|i| self.m[i] / (1.0 - 0.9_f32.powi(self.t as i32)));
        let v_hat: [f32; P] = array::from_fn(|i| self.v[i] / (1.0 - 0.999_f32.powi(self.t as i32)));

        // Update parameters
        let gradient = array::from_fn(|i| -learning_rate * m_hat[i] / (v_hat[i].sqrt() + 1e-8)); // epsilon = 1e-8

        let og_parameters = array::from_fn(|i| self.params[i].get_real());
        let new_params = (self.param_translator)(&og_parameters, &gradient);

        for (i, param) in new_params.iter().enumerate() {
            self.params[i].set_real(*param);
        }

        // Calculate the new cost
        let new_cost = (self.cost)(&new_params, &self.extra_data);

        if let Some(last_cost) = self.last_cost {
            if (last_cost - new_cost).abs() < self.cost_stagnation_threshold {
                // stagnation_is_happening

                if let Some(stagnation_value) = self.cost_stagnation_value {
                    if (new_cost - stagnation_value).abs() < self.cost_stagnation_threshold {
                        self.cost_stagnation_time += 1;
                    } else {
                        self.cost_stagnation_time = 0;
                        self.cost_stagnation_value = Some(new_cost);
                    }
                } else {
                    self.cost_stagnation_time = 0;
                    self.cost_stagnation_value = Some(new_cost);
                }
            } else {
                // reset_stagnation
                self.cost_stagnation_value = None;
                self.cost_stagnation_time = 0;
            }
        }

        self.last_cost = Some(new_cost);

        if VERBOSE {
            println!(
                "Gradient length: {gradient_size:?} - New cost: {} - Time: {}",
                self.last_cost.unwrap(),
                t0.elapsed().as_secs_f32()
            );
        }
    }

    fn get_model_params(&self) -> [f32; P] {
        self.params.clone().map(|e| e.get_real())
    }

    fn set_model_params(&mut self, parameters: [f32; P]) {
        let mut i = 0;
        self.params = parameters.map(|p| {
            i += 1;
            Dual::new_param(p, i - 1)
        });
    }

    fn found_local_minima(&self) -> bool {
        self.max_cost_stagnation_time < self.cost_stagnation_time
    }
}