burn 0.3.0

BURN: Burn Unstoppable Rusty Neurons
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

use burn_tensor::backend::Gradients;

use super::{load_state_gradients, register_state_gradients};
use crate as burn;
use crate::config::Config;
use crate::module::{ParamId, StateNamed};
use crate::tensor::backend::ADBackend;
use crate::tensor::{ElementConversion, Tensor};

/// Configuration to create momentum [Momentum](Momentum).
#[derive(Config)]
pub struct MomentumConfig {
    /// Momemtum factor
    #[config(default = 0.9)]
    pub momentum: f64,
    /// Dampening factor.
    #[config(default = 0.1)]
    pub dampening: f64,
    /// Enables Nesterov momentum, see [On the importance of initialization and momentum in deep learning](http://www.cs.toronto.edu/~hinton/absps/momentum.pdf).
    #[config(default = false)]
    pub nesterov: bool,
}

/// Momemtum implementation that transforms gradients.
pub struct Momentum<B: ADBackend> {
    momentum: B::Elem,
    dampening: f64,
    nesterov: bool,
    velocity: B::Gradients,
}

impl<B: ADBackend> Momentum<B> {
    pub fn new(config: &MomentumConfig) -> Self {
        Self {
            momentum: config.momentum.to_elem(),
            dampening: config.dampening,
            velocity: B::Gradients::empty(),
            nesterov: config.nesterov,
        }
    }

    pub fn transform<const D: usize>(
        &mut self,
        id: &ParamId,
        grad: Tensor<B::InnerBackend, D>,
    ) -> Tensor<B::InnerBackend, D> {
        let id = id.to_string();

        let velocity = match self.velocity.get::<D>(&id) {
            Some(grad_last_step) => grad
                .mul_scalar(1.0 - self.dampening)
                .add(&grad_last_step.mul_scalar(self.momentum)),
            None => grad.clone(),
        };

        // Update velocity
        self.velocity.register(id, velocity.clone());

        match self.nesterov {
            true => velocity.mul_scalar(self.momentum).add(&grad),
            false => velocity,
        }
    }

    pub fn register_state<const D: usize>(&self, id: &ParamId, state: &mut StateNamed<B::Elem>) {
        register_state_gradients::<D, B, _>(id, state, &self.velocity, Self::state_key);
    }

    pub fn load_state<const D: usize>(
        &mut self,
        id: &ParamId,
        state: &StateNamed<B::Elem>,
        device: &B::Device,
    ) {
        load_state_gradients::<D, B, _>(id, state, &mut self.velocity, Self::state_key, device);
    }

    fn state_key(id: &str) -> String {
        format!("momentum-{}", id)
    }
}