lumen-dataset 0.4.0

A tiny ML 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

use rand::seq::SliceRandom;
use super::Dataset;
use super::Batcher;

pub struct DataLoader<D, B> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item>
{
    dataset: D,
    batcher: B,
    batch_size: usize,
    shuffle: bool,
} 

pub struct DataLoaderIter<'a, D, B> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item>
{
    loader: &'a DataLoader<D, B>,
    cursor: usize,
    indices: Vec<usize>,
}

pub struct DataLoaderIntoIter<D, B> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item>
{
    loader: DataLoader<D, B>,
    cursor: usize,
    indices: Vec<usize>,
}

impl<D, B> DataLoader<D, B> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item> 
{
    pub fn new(dataset: D, batcher: B, batch_size: usize, shuffle: bool) -> Self {
        Self {
            dataset,
            batcher,
            batch_size,
            shuffle,
        }
    }

    pub fn dataset_len(&self) -> usize {
        self.dataset.len()
    }

    pub fn batch_count(&self) -> usize {
        self.dataset.len() / self.batch_size
    }

    pub fn batch_size(&self) -> usize {
        self.batch_size
    }

    pub fn iter<'a>(&'a self) -> DataLoaderIter<'a, D, B> {
        let indices = self.get_iter_indices();
        DataLoaderIter { loader: self, cursor: 0, indices }
    }

    fn get_iter_indices(&self) -> Vec<usize> {
        let len = self.dataset.len();
        let mut indices: Vec<usize> = (0..len).collect();
        
        if self.shuffle {
            let mut rng = rand::rng();
            indices.shuffle(&mut rng);
        }

        indices
    }
}

impl<D, B> DataLoader<D, B> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item> + Default
{
    pub fn from_dataset(dataset: D, batch_size: usize, shuffle: bool) -> Self {
        Self::new(dataset, B::default(), batch_size, shuffle)
    }
}

impl<D, B> Iterator for DataLoaderIntoIter<D, B> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item> 
{
    type Item = B::Output;
    fn next(&mut self) -> Option<Self::Item> {
        iter_next(&self.loader, &mut self.cursor, &self.indices)
    }
}

impl<'a, D, B> Iterator for DataLoaderIter<'a, D, B> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item> 
{
    type Item = B::Output;
    fn next(&mut self) -> Option<Self::Item> {
        iter_next(&self.loader, &mut self.cursor, &self.indices)
    }
}

impl<D, B> IntoIterator for DataLoader<D, B> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item> 
{
    type IntoIter = DataLoaderIntoIter<D, B>;
    type Item = B::Output;
    fn into_iter(self) -> Self::IntoIter {
        let indices = self.get_iter_indices();
        DataLoaderIntoIter {
            loader: self, 
            cursor: 0,
            indices,
        } 
    }
}

fn iter_next<D, B>(loader: &DataLoader<D, B>, cursor: &mut usize, indices: &[usize]) -> Option<B::Output> 
where 
    D: Dataset,
    B: Batcher<Item = D::Item> 
{
    let begin = *cursor;
    if begin >= loader.dataset.len() {
        None
    } else {
        let end = *cursor + loader.batch_size;
        let end = end.min(loader.dataset.len());
        let mut items = vec![];
        for index in begin..end {
            let index = indices[index];
            items.push(loader.dataset.get(index).unwrap());
        }        
        *cursor += items.len();
        let batch = loader.batcher.batch(items);
        Some(batch)
    }
}