concision_core/activate/impls/
impl_nonlinear.rs

1/*
2    Appellation: sigmoid <mod>
3    Contrib: FL03 <jo3mccain@icloud.com>
4*/
5use crate::activate::{ReLU, Sigmoid, Softmax, Tanh,
6    utils::sigmoid_derivative,
7};
8
9use ndarray::{Array, ArrayBase, Data, Dimension, ScalarOperand};
10use num_traits::{Float, One, Zero};
11
12impl<A, S, D> ReLU for ArrayBase<S, D>
13where
14    A: Copy + PartialOrd + Zero + One,
15    S: Data<Elem = A>,
16    D: Dimension,
17{
18    type Output = Array<A, D>;
19
20    fn relu(&self) -> Self::Output {
21        self.map(|&i| if i > A::zero() { i } else { A::zero() })
22    }
23
24    fn relu_derivative(&self) -> Self::Output {
25        self.map(|&i| if i > A::zero() { A::one() } else { A::zero() })
26    }
27}
28
29impl<A, S, D> Sigmoid for ArrayBase<S, D>
30where
31    A: Float + 'static,
32    S: Data<Elem = A>,
33    D: Dimension,
34{
35    type Output = Array<A, D>;
36
37    fn sigmoid(self) -> Self::Output {
38        let dim = self.dim();
39        let ones = Array::<A, D>::ones(dim);
40
41        (ones + self.map(|&i| i.neg().exp())).recip()
42    }
43
44    fn sigmoid_derivative(self) -> Self::Output {
45        self.mapv(|i| sigmoid_derivative(i))
46    }
47}
48
49impl<A, S, D> Softmax for ArrayBase<S, D>
50where
51    A: Float + ScalarOperand,
52    S: Data<Elem = A>,
53    D: Dimension,
54{
55    type Output = Array<A, D>;
56
57    fn softmax(&self) -> Self::Output {
58        let e = self.exp();
59        &e / e.sum()
60    }
61
62    fn softmax_derivative(&self) -> Self::Output {
63        let e = self.exp();
64        let sum = e.sum();
65        let softmax = &e / sum;
66
67        let ones = Array::<A, D>::ones(self.dim());
68        &softmax * (&ones - &softmax)
69    }
70}
71
72impl<A, S, D> Tanh for ArrayBase<S, D>
73where
74    A: Float + 'static,
75    S: Data<Elem = A>,
76    D: Dimension,
77{
78    type Output = Array<A, D>;
79
80    fn tanh(&self) -> Self::Output {
81        self.map(|i| i.tanh())
82    }
83
84    fn tanh_derivative(&self) -> Self::Output {
85        self.map(|i| A::one() - i.tanh().powi(2))
86    }
87}
concision_core/activate/impls/impl_nonlinear.rs

concision_core/activate/impls/
impl_nonlinear.rs
