concision_core/activate/impls/
impl_nonlinear.rs

1/*
2    Appellation: sigmoid <mod>
3    Contrib: FL03 <jo3mccain@icloud.com>
4*/
5use crate::{
6    activate::{ReLU, Sigmoid, Softmax, Tanh},
7    sigmoid_derivative,
8};
9
10use ndarray::{Array, ArrayBase, Data, Dimension, ScalarOperand};
11use num_traits::{Float, One, Zero};
12
13impl<A, S, D> ReLU for ArrayBase<S, D>
14where
15    A: Copy + PartialOrd + Zero + One,
16    S: Data<Elem = A>,
17    D: Dimension,
18{
19    type Output = Array<A, D>;
20
21    fn relu(&self) -> Self::Output {
22        self.map(|&i| if i > A::zero() { i } else { A::zero() })
23    }
24
25    fn relu_derivative(&self) -> Self::Output {
26        self.map(|&i| if i > A::zero() { A::one() } else { A::zero() })
27    }
28}
29
30impl<A, S, D> Sigmoid for ArrayBase<S, D>
31where
32    A: Float + 'static,
33    S: Data<Elem = A>,
34    D: Dimension,
35{
36    type Output = Array<A, D>;
37
38    fn sigmoid(&self) -> Self::Output {
39        let dim = self.dim();
40        let ones = Array::<A, D>::ones(dim);
41
42        (ones + self.map(|&i| i.neg().exp())).recip()
43    }
44
45    fn sigmoid_derivative(&self) -> Self::Output {
46        self.mapv(|i| sigmoid_derivative(i))
47    }
48}
49
50impl<A, S, D> Softmax for ArrayBase<S, D>
51where
52    A: Float + ScalarOperand,
53    S: Data<Elem = A>,
54    D: Dimension,
55{
56    type Output = Array<A, D>;
57
58    fn softmax(&self) -> Self::Output {
59        let e = self.exp();
60        &e / e.sum()
61    }
62
63    fn softmax_derivative(&self) -> Self::Output {
64        let e = self.exp();
65        let sum = e.sum();
66        let softmax = &e / sum;
67
68        let ones = Array::<A, D>::ones(self.dim());
69        &softmax * (&ones - &softmax)
70    }
71}
72
73impl<A, S, D> Tanh for ArrayBase<S, D>
74where
75    A: Float + 'static,
76    S: Data<Elem = A>,
77    D: Dimension,
78{
79    type Output = Array<A, D>;
80
81    fn tanh(&self) -> Self::Output {
82        self.map(|i| i.tanh())
83    }
84
85    fn tanh_derivative(&self) -> Self::Output {
86        self.map(|i| A::one() - i.tanh().powi(2))
87    }
88}
concision_core/activate/impls/impl_nonlinear.rs

concision_core/activate/impls/
impl_nonlinear.rs
