Struct ducky_learn::layers::Dense1d

pub struct Dense1d { /* private fields */ }

Implementations

impl Dense1d

pub fn from( activation: fn(_: Array1<f64>) -> Array1<f64>, deriv_activation: fn(_: Array1<f64>) -> Array1<f64>, weights: Array2<f64>, bias: Array1<f64> ) -> Self

Create Dense1d layer with full control over every part of the layer

Arguments

• activation: Activation function of whole 1d array
• weights: 2d array that has to be of shape( output, input )
• bias: 1d array of basis that has to be the size of the output

returns: Dense1d

Examples

use ducky_learn::layers::*;
use ndarray::{arr1, arr2};

let layer = Dense1d::from(
                |x| x, // Activation function that is does nothing
                |x| x.map(|i| 1f64), // Derivative of Activation function
                arr2(&[[1., 1.], [1., 1.]]), // 2x2 array
                arr1(&[1., 1.]) // len 2
            );

pub fn new( input_size: usize, layer_size: usize, activation_fn: fn(_: Array1<f64>) -> Array1<f64>, deriv_activation_fn: fn(_: Array1<f64>) -> Array1<f64> ) -> Self

Create randomly set weights and bias’s for the dense1d layer. Creates weights and bias’s using a normal distribution from -1. -> 1.

Arguments

• input_size: size of input array
• layer_size: number of nodes in the layer
• activation_fn: activation function for the layer

returns: Dense1d

Examples

use ducky_learn::layers::*;
use ndarray::{arr1, arr2};

let layer = Dense1d::new(5, 10, |x| x, |x| x);
let input_array = arr1(&[
    1., 1., 1., 1., 1.
]);

layer.pass(input_array);

Trait Implementations

impl Layer1d for Dense1d

fn pass(&self, input_array: Array1<f64>) -> (Array1<f64>, Array1<f64>)

Feeds forward the 1d array through the layer.