burn-optim 0.22.0-pre.1

Optimizer building blocks for the Burn deep learning framework
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

use burn_core::{self as burn};

use super::cosine::CosineAnnealingLrSchedulerConfig;
use super::exponential::ExponentialLrSchedulerConfig;
use super::linear::LinearLrSchedulerConfig;
use super::noam::NoamLrSchedulerConfig;
use super::{LrScheduler, LrSchedulerRecord, String};
use crate::LearningRate;
use crate::lr_scheduler::module_lr_scheduler::ModuleLrScheduler;
use crate::lr_scheduler::step::StepLrSchedulerConfig;
use crate::lr_scheduler::{DynLrScheduler, LrSchedulerConfig};

use burn::config::Config;

/// Compose multiple [learning rate schedulers](LrScheduler) together.
#[derive(Config, Debug)]
pub struct ComposedLrSchedulerConfig {
    #[config(default = "Vec::new()")]
    schedulers: Vec<LrSchedulerConfig>,
    #[config(default = "SchedulerReduction::Prod")]
    reduction: SchedulerReduction,
}

/// Compose multiple [learning rate schedulers](LrScheduler) together.
#[derive(Clone)]
pub struct ComposedLrScheduler {
    schedulers: Vec<DynLrScheduler>,
    reduction: SchedulerReduction,
}

/// Defines how the learning rates generated by the schedulers are combined.
#[derive(Config, Debug, Copy)]
pub enum SchedulerReduction {
    /// All learning rates are averaged.
    Avg,
    /// All learning rates are summed.
    Sum,
    /// All learning rates are multiplied.
    Prod,
}

impl ComposedLrSchedulerConfig {
    /// Initialize the learning rate scheduler.
    pub(crate) fn build(&self) -> Result<ComposedLrScheduler, String> {
        let mut schedulers: Vec<DynLrScheduler> = Vec::with_capacity(self.schedulers.len());
        for config in self.schedulers.iter() {
            schedulers.push(config.build()?);
        }

        Ok(ComposedLrScheduler {
            schedulers,
            reduction: self.reduction,
        })
    }

    /// Initializes a [module learning rate scheduler](ModuleLrScheduler).
    pub fn init(&self) -> Result<ModuleLrScheduler, String> {
        self.build().map(|s| s.into())
    }

    /// Appends a [constant learning rate](crate::lr_scheduler::constant::ConstantLr).
    pub fn constant(mut self, lr: LearningRate) -> Self {
        self.schedulers.push(LrSchedulerConfig::Constant(lr));
        self
    }

    /// Appends a [linear scheduler](crate::lr_scheduler::linear::LinearLrScheduler).
    pub fn linear(mut self, config: LinearLrSchedulerConfig) -> Self {
        self.schedulers.push(LrSchedulerConfig::Linear(config));
        self
    }

    /// Appends a [cosine scheduler](ComposedLrSchedulerConfig).
    pub fn cosine(mut self, config: CosineAnnealingLrSchedulerConfig) -> Self {
        self.schedulers.push(LrSchedulerConfig::Cosine(config));
        self
    }

    /// Appends an [exponential scheduler](crate::lr_scheduler::exponential::ExponentialLrScheduler).
    pub fn exponential(mut self, config: ExponentialLrSchedulerConfig) -> Self {
        self.schedulers.push(LrSchedulerConfig::Exponential(config));
        self
    }

    /// Appends a [noam scheduler](crate::lr_scheduler::noam::NoamLrScheduler).
    pub fn noam(mut self, config: NoamLrSchedulerConfig) -> Self {
        self.schedulers.push(LrSchedulerConfig::Noam(config));
        self
    }

    /// Appends a [step scheduler](crate::lr_scheduler::step::StepLrScheduler).
    pub fn step(mut self, config: StepLrSchedulerConfig) -> Self {
        self.schedulers.push(LrSchedulerConfig::Step(config));
        self
    }

    /// Appends a [composed scheduler](ComposedLrScheduler).
    pub fn composed(mut self, config: Self) -> Self {
        self.schedulers.push(LrSchedulerConfig::Composed(config));
        self
    }
}

impl ComposedLrScheduler {
    /// Add a custom learning rate scheduler to existing ones.
    pub fn with_custom_scheduler<S: LrScheduler + 'static>(mut self, scheduler: S) -> Self {
        self.schedulers.push(scheduler.into());
        self
    }
}

impl LrScheduler for ComposedLrScheduler {
    fn step(&mut self) -> LearningRate {
        let mut step = match self.reduction {
            SchedulerReduction::Avg => 0.0,
            SchedulerReduction::Sum => 0.0,
            SchedulerReduction::Prod => 1.0,
        };
        let num_scheduler = self.schedulers.len() as f64;

        for lr in self.schedulers.iter_mut().map(|s| s.step()) {
            step = match self.reduction {
                SchedulerReduction::Avg => step + (lr / num_scheduler),
                SchedulerReduction::Sum => step + lr,
                SchedulerReduction::Prod => step * lr,
            }
        }

        step
    }

    fn to_record(&self) -> LrSchedulerRecord {
        let mut record = LrSchedulerRecord::new();
        for (index, item) in self.schedulers.iter().enumerate() {
            let sub = item.to_record();
            record = record.with_record(&index.to_string(), sub);
        }
        record
    }

    fn load_record(&mut self, record: LrSchedulerRecord) {
        self.schedulers = self
            .schedulers
            .clone()
            .iter_mut()
            .enumerate()
            .map(|(index, item)| {
                let sub = record.record(&index.to_string());
                item.clone().load_record(sub)
            })
            .collect();
    }
}