newron 0.5.1

A Rust library to train and infer deep learning models.
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

use crate::{tensor::Tensor, loss::loss::Loss, utils, layers::softmax::Softmax};
pub struct CategoricalEntropy {}

impl Loss for CategoricalEntropy {
    fn compute_loss(&self, y_true: &Tensor, y_pred: &Tensor) -> f64 {
        let m = y_true.shape[0];

        let indices = utils::one_hot_encoded_tensor_to_indices(y_true);

        // Even if last layer is Softmax, we softmax-it again for numerical stability
        let softmax_value = Softmax::softmax(y_pred);

        let mut p = Vec::new();
        for (row, indice) in indices.iter().enumerate() {
            p.push(softmax_value.get_value(row, *indice));
        }

        let log_likelihood: Vec<f64> = p.iter().map(|x| -(x.ln())).collect();
        log_likelihood.iter().sum::<f64>() / m as f64
    }

    fn compute_loss_grad(&self, y_true: &Tensor, y_pred: &Tensor) -> Tensor {
        let rows = y_true.shape[0];
        let cols = y_true.shape[1];

        let indices = utils::one_hot_encoded_tensor_to_indices(y_true);

        // Even if last layer is Softmax, we softmax-it again for numerical stability
        let softmax_value = Softmax::softmax(y_pred);

        let mut data: Vec<f64> = Vec::new();

        for row in 0..rows {
            for col in 0..cols {
                let indice = indices[row];
                let mut value = softmax_value.get_value(row, col);
                if col == indice { value -= 1.0 }
                value /= rows as f64;
                data.push(value);
            }
        }

        Tensor {
            data,
            shape: vec![rows, cols]
        }

    }
}