concision_core/activate/impls/
impl_activate_nonlinear.rs

1/*
2    Appellation: impl_nonlinear <module>
3    Created At: 2025.12.14:11:13:15
4    Contrib: @FL03
5*/
6use crate::activate::{ReLUActivation, SigmoidActivation, SoftmaxActivation, TanhActivation};
7use ndarray::{Array, ArrayBase, Data, Dimension, ScalarOperand};
8use num_traits::{Float, One, Zero};
9
10impl<A, S, D> ReLUActivation for ArrayBase<S, D, A>
11where
12    A: Copy + PartialOrd + Zero + One,
13    S: Data<Elem = A>,
14    D: Dimension,
15{
16    type Output = Array<A, D>;
17
18    fn relu(&self) -> Self::Output {
19        self.map(|&i| if i > A::zero() { i } else { A::zero() })
20    }
21
22    fn relu_derivative(&self) -> Self::Output {
23        self.map(|&i| if i > A::zero() { A::one() } else { A::zero() })
24    }
25}
26
27impl<A, S, D> SigmoidActivation for ArrayBase<S, D, A>
28where
29    A: 'static + Float,
30    S: Data<Elem = A>,
31    D: Dimension,
32{
33    type Output = Array<A, D>;
34
35    fn sigmoid(self) -> Self::Output {
36        let dim = self.dim();
37        let ones = Array::<A, D>::ones(dim);
38
39        (ones + self.signum().exp()).recip()
40    }
41
42    fn sigmoid_derivative(self) -> Self::Output {
43        self.mapv(|i| {
44            let s = (A::one() + i.neg().exp()).recip();
45            s * (A::one() - s)
46        })
47    }
48}
49
50impl<A, S, D> SoftmaxActivation for ArrayBase<S, D, A>
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 exp = self.exp();
60        &exp / exp.sum()
61    }
62
63    fn softmax_derivative(&self) -> Self::Output {
64        let softmax = self.softmax();
65
66        let ones = Array::<A, D>::ones(self.dim());
67        &softmax * (&ones - &softmax)
68    }
69}
70
71impl<A, S, D> TanhActivation for ArrayBase<S, D, A>
72where
73    A: 'static + Float,
74    S: Data<Elem = A>,
75    D: Dimension,
76{
77    type Output = Array<A, D>;
78
79    fn tanh(&self) -> Self::Output {
80        self.mapv(|i| i.tanh())
81    }
82
83    fn tanh_derivative(&self) -> Self::Output {
84        self.mapv(|i| A::one() - i.tanh().powi(2))
85    }
86}
concision_core/activate/impls/impl_activate_nonlinear.rs

concision_core/activate/impls/
impl_activate_nonlinear.rs
