rusty_machine/learning/toolkit/
activ_fn.rs

1//! Activation Functions.
2//!
3//! This module contains a number of structs implementing the `ActivationFunc` trait.
4//!
5//! These structs are used within Neural Networks and
6//! Generalized Linear Regression (not yet implemented).
7//!
8//! You can also create your own custom activation Functions for use in your models.
9//! Just create a unit struct implementing the `ActivationFunc` trait.
10
11/// Trait for activation functions in models.
12pub trait ActivationFunc {
13    /// The activation function.
14    fn func(x: f64) -> f64;
15
16    /// The gradient of the activation function.
17    fn func_grad(x: f64) -> f64;
18
19    /// The inverse of the activation function.
20    fn func_inv(x: f64) -> f64;
21}
22
23/// Sigmoid activation function.
24#[derive(Clone, Copy, Debug)]
25pub struct Sigmoid;
26
27impl ActivationFunc for Sigmoid {
28    /// Sigmoid function.
29    ///
30    /// Returns 1 / ( 1 + e^-t).
31    fn func(x: f64) -> f64 {
32        1.0 / (1.0 + (-x).exp())
33    }
34
35    /// Gradient of sigmoid function.
36    ///
37    /// Evaluates to (1 - e^-t) / (1 + e^-t)^2
38    fn func_grad(x: f64) -> f64 {
39        Self::func(x) * (1f64 - Self::func(x))
40    }
41
42    fn func_inv(x: f64) -> f64 {
43        (x / (1f64 - x)).ln()
44    }
45}
46
47/// Linear activation function.
48#[derive(Clone, Copy, Debug)]
49pub struct Linear;
50
51impl ActivationFunc for Linear {
52    fn func(x: f64) -> f64 {
53        x
54    }
55
56    fn func_grad(_: f64) -> f64 {
57        1f64
58    }
59
60    fn func_inv(x: f64) -> f64 {
61        x
62    }
63}
64
65/// Exponential activation function.
66#[derive(Clone, Copy, Debug)]
67pub struct Exp;
68
69impl ActivationFunc for Exp {
70    fn func(x: f64) -> f64 {
71        x.exp()
72    }
73
74    fn func_grad(x: f64) -> f64 {
75        Self::func(x)
76    }
77
78    fn func_inv(x: f64) -> f64 {
79        x.ln()
80    }
81}
rusty_machine/learning/toolkit/activ_fn.rs

rusty_machine/learning/toolkit/
activ_fn.rs
