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

use crate::metric::{
    AccuracyInput, Adaptor, ConfusionStatsInput, HammingScoreInput, LossInput, PerplexityInput,
    TopKAccuracyInput, processor::ItemLazy,
};
use burn_core::tensor::{Device, Int, Tensor, Transaction};

/// Simple classification output adapted for multiple metrics.
///
/// Supported metrics:
/// - Accuracy
/// - AUROC
/// - TopKAccuracy
/// - Perplexity
/// - Precision (via ConfusionStatsInput)
/// - Recall (via ConfusionStatsInput)
/// - FBetaScore (via ConfusionStatsInput)
/// - Loss.
#[derive(new)]
pub struct ClassificationOutput {
    /// The loss.
    pub loss: Tensor<1>,

    /// The class logits or probabilities. Shape: \[batch_size, num_classes\].
    pub output: Tensor<2>,

    /// The ground truth class index for each sample. Shape: \[batch_size\].
    pub targets: Tensor<1, Int>,
}

impl ItemLazy for ClassificationOutput {
    fn sync(self) -> Self {
        let [output, loss, targets] = Transaction::default()
            .register(self.output)
            .register(self.loss)
            .register(self.targets)
            .execute()
            .try_into()
            .expect("Correct amount of tensor data");

        let device: Device = Device::flex();

        ClassificationOutput {
            output: Tensor::from_data(output, &device),
            loss: Tensor::from_data(loss, &device),
            targets: Tensor::from_data(targets, &device),
        }
    }
}

impl Adaptor<AccuracyInput> for ClassificationOutput {
    fn adapt(&self) -> AccuracyInput {
        AccuracyInput::new(self.output.clone(), self.targets.clone())
    }
}

impl Adaptor<LossInput> for ClassificationOutput {
    fn adapt(&self) -> LossInput {
        LossInput::new(self.loss.clone())
    }
}

impl Adaptor<TopKAccuracyInput> for ClassificationOutput {
    fn adapt(&self) -> TopKAccuracyInput {
        TopKAccuracyInput::new(self.output.clone(), self.targets.clone())
    }
}

impl Adaptor<PerplexityInput> for ClassificationOutput {
    fn adapt(&self) -> PerplexityInput {
        PerplexityInput::new(self.output.clone(), self.targets.clone())
    }
}

impl Adaptor<ConfusionStatsInput> for ClassificationOutput {
    fn adapt(&self) -> ConfusionStatsInput {
        let [_, num_classes] = self.output.dims();
        if num_classes > 1 {
            ConfusionStatsInput::new(
                self.output.clone(),
                self.targets.clone().one_hot(num_classes).bool(),
            )
        } else {
            ConfusionStatsInput::new(
                self.output.clone(),
                self.targets.clone().unsqueeze_dim(1).bool(),
            )
        }
    }
}

/// Multi-label classification output adapted for multiple metrics.
///
/// Supported metrics:
/// - HammingScore
/// - Precision (via ConfusionStatsInput)
/// - Recall (via ConfusionStatsInput)
/// - FBetaScore (via ConfusionStatsInput)
/// - Loss
#[derive(new)]
pub struct MultiLabelClassificationOutput {
    /// The loss.
    pub loss: Tensor<1>,

    /// The label logits or probabilities. Shape: \[batch_size, num_classes\].
    pub output: Tensor<2>,

    /// The ground truth labels. Shape: \[batch_size, num_classes\].
    pub targets: Tensor<2, Int>,
}

impl ItemLazy for MultiLabelClassificationOutput {
    fn sync(self) -> Self {
        let [output, loss, targets] = Transaction::default()
            .register(self.output)
            .register(self.loss)
            .register(self.targets)
            .execute()
            .try_into()
            .expect("Correct amount of tensor data");

        let device: Device = Device::flex();

        MultiLabelClassificationOutput {
            output: Tensor::from_data(output, &device),
            loss: Tensor::from_data(loss, &device),
            targets: Tensor::from_data(targets, &device),
        }
    }
}

impl Adaptor<HammingScoreInput> for MultiLabelClassificationOutput {
    fn adapt(&self) -> HammingScoreInput {
        HammingScoreInput::new(self.output.clone(), self.targets.clone())
    }
}

impl Adaptor<LossInput> for MultiLabelClassificationOutput {
    fn adapt(&self) -> LossInput {
        LossInput::new(self.loss.clone())
    }
}

impl Adaptor<ConfusionStatsInput> for MultiLabelClassificationOutput {
    fn adapt(&self) -> ConfusionStatsInput {
        ConfusionStatsInput::new(self.output.clone(), self.targets.clone().bool())
    }
}