tract-core 0.3.2

Tiny, no-nonsense, self contained, TensorFlow and ONNX inference
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

element_map!(Tanh, [f32], tanh_f32);

const LOW: f32 = -9.0;
const HIGH: f32 = 9.0;
const ALPHA_13: f32 = -2.76076847742355e-16;
const ALPHA_11: f32 = 2.00018790482477e-13;
const ALPHA_9: f32 = -8.60467152213735e-11;
const ALPHA_7: f32 = 5.12229709037114e-08;
const ALPHA_5: f32 = 1.48572235717979e-05;
const ALPHA_3: f32 = 6.37261928875436e-04;
const ALPHA_1: f32 = 4.89352455891786e-03;
const BETA_6: f32 = 1.19825839466702e-06;
const BETA_4: f32 = 1.18534705686654e-04;
const BETA_2: f32 = 2.26843463243900e-03;
const BETA_0: f32 = 4.89352518554385e-03;

pub fn tanh_f32(x: f32) -> f32 {
    if x <= LOW {
        return -1.0;
    }
    if x >= HIGH {
        return 1.0;
    }
    let x2 = x * x;

    let p = x2 * ALPHA_13 + ALPHA_11;
    let p = x2 * p + ALPHA_9;
    let p = x2 * p + ALPHA_7;
    let p = x2 * p + ALPHA_5;
    let p = x2 * p + ALPHA_3;
    let p = x2 * p + ALPHA_1;
    let p = p * x;

    let q = x2 * BETA_6 + BETA_4;
    let q = x2 * q + BETA_2;
    let q = x2 * q + BETA_0;

    p / q
}