burn-train 0.22.0-pre.1

Training crate for the Burn 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
154
155
156
157

use super::MetricMetadata;
use super::state::{FormatOptions, NumericMetricState};
use crate::metric::{Metric, MetricAttributes, MetricName, Numeric, SerializedEntry};
use burn_core::tensor::{Int, Tensor};

/// The accuracy metric.
#[derive(Clone)]
pub struct AccuracyMetric {
    name: MetricName,
    state: NumericMetricState,
    pad_token: Option<usize>,
}

/// The [accuracy metric](AccuracyMetric) input type.
#[derive(new)]
pub struct AccuracyInput {
    outputs: Tensor<2>,
    targets: Tensor<1, Int>,
}

impl Default for AccuracyMetric {
    fn default() -> Self {
        Self::new()
    }
}

impl AccuracyMetric {
    /// Creates the metric.
    pub fn new() -> Self {
        Self {
            name: MetricName::new("Accuracy".to_string()),
            state: Default::default(),
            pad_token: Default::default(),
        }
    }

    /// Sets the pad token.
    pub fn with_pad_token(mut self, index: usize) -> Self {
        self.pad_token = Some(index);
        self
    }
}

impl Metric for AccuracyMetric {
    type Input = AccuracyInput;

    fn update(&mut self, input: &AccuracyInput, _metadata: &MetricMetadata) -> SerializedEntry {
        let targets = input.targets.clone();
        let outputs = input.outputs.clone();

        let [batch_size, _n_classes] = outputs.dims();

        let outputs = outputs.argmax(1).reshape([batch_size]);

        let accuracy = match self.pad_token {
            Some(pad_token) => {
                let mask = targets.clone().equal_scalar(pad_token as i64);
                let matches = outputs.equal(targets).float().mask_fill(mask.clone(), 0);
                let num_pad = mask.float().sum();

                let acc = matches.sum() / (num_pad.neg() + batch_size as f32);

                acc.into_scalar::<f64>()
            }
            None => outputs.equal(targets).int().sum().into_scalar::<f64>() / batch_size as f64,
        };

        self.state.update(100.0 * accuracy, batch_size);
        self.state
            .compute_update(FormatOptions::new(self.name()).unit("%").precision(2))
    }

    fn compute(&mut self) -> SerializedEntry {
        self.state
            .compute_final(FormatOptions::new(self.name()).unit("%").precision(2))
    }

    fn clear(&mut self) {
        self.state.reset()
    }

    fn name(&self) -> MetricName {
        self.name.clone()
    }

    fn attributes(&self) -> MetricAttributes {
        super::NumericAttributes {
            unit: Some("%".to_string()),
            higher_is_better: true,
        }
        .into()
    }
}

impl Numeric for AccuracyMetric {
    fn value(&self) -> Option<super::NumericEntry> {
        Some(self.state.current_value())
    }

    fn running_value(&self) -> Option<super::NumericEntry> {
        Some(self.state.running_value())
    }

    fn final_value(&self) -> super::NumericEntry {
        self.state.final_value()
    }
}

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

    #[test]
    fn test_accuracy_without_padding() {
        let device = Default::default();
        let mut metric = AccuracyMetric::new();
        let input = AccuracyInput::new(
            Tensor::from_data(
                [
                    [0.0, 0.2, 0.8], // 2
                    [1.0, 2.0, 0.5], // 1
                    [0.4, 0.1, 0.2], // 0
                    [0.6, 0.7, 0.2], // 1
                ],
                &device,
            ),
            Tensor::from_data([2, 2, 1, 1], &device),
        );

        let _entry = metric.update(&input, &MetricMetadata::fake());
        assert_eq!(50.0, metric.value().unwrap().current());
    }

    #[test]
    fn test_accuracy_with_padding() {
        let device = Default::default();
        let mut metric = AccuracyMetric::new().with_pad_token(3);
        let input = AccuracyInput::new(
            Tensor::from_data(
                [
                    [0.0, 0.2, 0.8, 0.0], // 2
                    [1.0, 2.0, 0.5, 0.0], // 1
                    [0.4, 0.1, 0.2, 0.0], // 0
                    [0.6, 0.7, 0.2, 0.0], // 1
                    [0.0, 0.1, 0.2, 5.0], // Predicted padding should not count
                    [0.0, 0.1, 0.2, 0.0], // Error on padding should not count
                    [0.6, 0.0, 0.2, 0.0], // Error on padding should not count
                ],
                &device,
            ),
            Tensor::from_data([2, 2, 1, 1, 3, 3, 3], &device),
        );

        let _entry = metric.update(&input, &MetricMetadata::fake());
        assert_eq!(50.0, metric.value().unwrap().current());
    }
}