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

//! This crate achieves the functionality of [sklearn's train_test_split](https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.train_test_split.html)
//! to generate splits of the data (in this case, a slice), generalized for an
//! arbitrary number of splits. It
//! both provides functions (see [split_parts](https://docs.rs/rand_split/0.2.0/rand_split/fn.split_parts.html)) that work on slices and iterator traits (see [PartsSplit](https://docs.rs/rand_split/0.2.0/rand_split/trait.PartsSplit.html))to work
//! with streams of data.
//!
//! Check out the [examples in the repository](https://github.com/carrascomj/rand_split/tree/trunk/examples)
//! for more information.
use rand::prelude::*;

mod stream;
pub use stream::PartsSplit;

mod stream_ttv;
pub use stream_ttv::TTVSplit;

/// Split the elements of a container in randomized sets which contain a
/// a part (in `splits`) of the input.
///
/// # Example
///
/// ```
/// use rand_split::split_parts;
///
/// println!("{:#?}", split_parts(&mut [1,2,3,4,5,6,8,9,10], &[0.4, 0.2, 0.4]));
/// ```
pub fn split_parts<'a, T>(cont: &'a mut [T], splits: &[f32]) -> Vec<&'a mut [T]>
where
    T: Clone,
{
    let n = cont.len();
    let n_weights = splits.len();
    let total_weights = splits.iter().sum::<f32>();

    let mut out = Vec::with_capacity(n);
    let mut left = cont;
    left.shuffle(&mut rand::thread_rng());
    for sp in splits
        .iter()
        .map(|w| (w * (n as f32) / total_weights) as usize)
        .take(n_weights - 1)
    {
        let (right, l) = left.split_at_mut(sp);
        out.push(right);
        left = l;
    }
    out.push(left);

    out
}

/// Generate train-test splits. Wrapper around [`split_parts`](./split_parts)
/// # Examples
///
/// ```
/// use rand_split::train_test_split;
///
/// let mut cont = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13];
/// let total_len = cont.len();
/// let result = train_test_split(&mut cont, 0.8, 0.2);
/// assert_eq!(
///     result.iter().map(|inner| inner.len()).sum::<usize>(),
///     total_len
/// );
/// ```
pub fn train_test_split<T>(cont: &mut [T], train: f32, test: f32) -> Vec<&mut [T]>
where
    T: Clone,
{
    split_parts(cont, &[train, test])
}

/// Generate train-test-validation splits. Wrapper around [`split_parts`](./split_parts)
/// # Examples
///
/// ```
/// use rand_split::ttv_split;
///
/// let mut cont = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13];
/// let total_len = cont.len();
/// let result = ttv_split(&mut cont, 0.6, 0.2, 0.2);
/// assert_eq!(
///     result.iter().map(|inner| inner.len()).sum::<usize>(),
///     total_len
/// );
/// ```
pub fn ttv_split<T>(cont: &mut [T], train: f32, test: f32, validation: f32) -> Vec<&mut [T]>
where
    T: Clone,
{
    split_parts(cont, &[train, test, validation])
}

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

    #[test]
    fn correct_split_sizes_even_number() {
        let splits = [0.2, 0.7, 0.1];
        let mut cont = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
        let total_len = cont.len();
        let result = split_parts(&mut cont, &splits);
        assert_eq!(
            result.iter().map(|inner| inner.len()).sum::<usize>(),
            total_len
        );
        assert_eq!(result.len(), splits.len());
    }

    #[test]
    fn correct_split_sizes_odd_number() {
        let splits = [0.2, 0.2, 0.3, 0.3];
        let mut cont = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13];
        let total_len = cont.len();
        let result = split_parts(&mut cont, &splits);
        assert_eq!(
            result.iter().map(|inner| inner.len()).sum::<usize>(),
            total_len
        );
        assert_eq!(result.len(), splits.len());
    }

    #[test]
    fn unnormalized_weights_is_approx_correct() {
        let splits = [2., 2., 3., 10.];
        let mut cont = (0..1999).collect::<Vec<usize>>();
        let result = split_parts(&mut cont, &splits);
        assert!(result[3].len() > result[0..2].iter().map(|sp| sp.len()).sum());
        assert!(result[1].len() < result[2].len());
    }

    #[test]
    fn unnormalized_weights_preserve_data() {
        let splits = [2., 2., 3., 6.];
        let mut cont = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13];
        let total_len = cont.len();
        let result = split_parts(&mut cont, &splits);
        assert_eq!(
            result.iter().map(|inner| inner.len()).sum::<usize>(),
            total_len
        );
        assert_eq!(result.len(), splits.len());
    }
}