kaffe 0.2.0

Pytorch, but make it rust :)
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

//!  Pooling
#![warn(missing_docs)]
use std::{error::Error, str::FromStr};

use crate::{Tensor, TensorElement};
use rayon::prelude::*;

/// Max Pooling
///
/// # Examples
///
/// ```
/// use kaffe::Tensor;
/// use kaffe::nn::pooling::MaxPool;
///
/// let matrix = Tensor::init(2.0, vec![4,4]);
///
/// let res = MaxPool(&matrix, 2, 0);
///
/// assert_eq!(res.shape, vec![2,2]);
///
/// ```
pub fn MaxPool<'a, T>(x: &Tensor<'a, T>, stride: usize, padding: usize) -> Tensor<'a, T>
where
    T: TensorElement,
    <T as FromStr>::Err: Error,
{
    let pred = |slice: &[T]| {
        slice
            .par_iter()
            .cloned()
            .max_by(|a, b| a.partial_cmp(b).unwrap())
    };

    pool(x, stride, padding, pred)
}

/// Min Pooling
///
/// # Examples
///
/// ```
/// use kaffe::Tensor;
/// use kaffe::nn::pooling::MinPool;
///
/// let mut matrix = Tensor::init(2.0, vec![4,4]);
///
/// matrix.set(vec![0,0], -2.0);
/// matrix.set(vec![0,3], -2.0);
/// matrix.set(vec![3,0], -2.0);
/// matrix.set(vec![3,3], -2.0);
///
/// let res = MinPool(&matrix, 2, 0);
///
/// assert_eq!(res.shape, vec![2,2]);
/// assert_eq!(res.get(vec![0,0]).unwrap(), -2.0);
/// assert_eq!(res.get(vec![0,1]).unwrap(), -2.0);
/// assert_eq!(res.get(vec![1,0]).unwrap(), -2.0);
/// assert_eq!(res.get(vec![1,1]).unwrap(), -2.0);
///
/// ```
pub fn MinPool<'a, T>(x: &Tensor<'a, T>, stride: usize, padding: usize) -> Tensor<'a, T>
where
    T: TensorElement,
    <T as FromStr>::Err: Error,
{
    let pred = |slice: &[T]| {
        slice
            .par_iter()
            .cloned()
            .min_by(|a, b| a.partial_cmp(b).unwrap())
    };

    pool(x, stride, padding, pred)
}

/// Avg Pooling
///
/// # Examples
///
/// ```
/// use kaffe::Tensor;
/// use kaffe::nn::pooling::AvgPool;
///
/// let matrix = Tensor::init(2.0, vec![4,4]);
///
/// let res = AvgPool(&matrix, 2, 0);
///
/// assert_eq!(res.shape, vec![2,2]);
/// //assert_eq!(res.get(vec![0,0]), 2.0);
///
/// ```
pub fn AvgPool<'a, T>(x: &Tensor<'a, T>, stride: usize, padding: usize) -> Tensor<'a, T>
where
    T: TensorElement,
    <T as FromStr>::Err: Error,
{
    let average =
        |slice: &[T]| Some(slice.par_iter().cloned().sum::<T>() / T::from(slice.len()).unwrap());

    pool(x, stride, padding, average)
}

/// Internal Helper function that actually performs the
/// predicates given, so pooling becomes easier
fn pool<'a, T, F>(x: &Tensor<'a, T>, stride: usize, padding: usize, mut pred: F) -> Tensor<'a, T>
where
    F: FnMut(&[T]) -> Option<T> + Send + Sync + 'static,
    T: TensorElement,
    <T as FromStr>::Err: Error,
{
    // Calculate the dimensions of the resulting matrix
    let out_rows = ((x.shape.iter().nth_back(1).unwrap() + 2 * padding - stride) / stride) + 1;
    let out_cols = ((x.shape.iter().nth_back(0).unwrap() + 2 * padding - stride) / stride) + 1;

    let mut pooled = Tensor::zeros(vec![out_rows, out_cols]);

    // Perform the pooling operation
    for i in 0..out_rows {
        for j in 0..out_cols {
            // Calculate the start indices of the pooling window
            let start_row = i * stride;
            let start_col = j * stride;

            // Extract the slice from the original matrix
            let slice = x.get_vec_slice(vec![start_row, start_col], stride, stride);

            // Apply the predicate to the slice and get the result
            if let Some(value) = pred(&slice) {
                pooled.set(vec![i, j], value);
            }
        }
    }

    pooled
}