Function gradient_descent

pub fn gradient_descent(
    x: &Tensor<f64>,
    y: &Tensor<f64>,
    w: &Tensor<f64>,
    l: f64,
    logistic: bool,
) -> Tensor<f64>

👎Deprecated since 0.4.0

The gradient_descent function performs a single step of the gradient descent optimization algorithm.

Arguments

• x - All values of all features arranged in a m * n matrix, where m is nuber of data points and n is number of features.
• w - The current value of the weights for each feature, arranged in n * 1 tensor, where n is number of features.
• y - The collected target values arranged in a m * 1 matrix, where m is the number of data points.
• l - The learning rate, which controls the size of the update step.
• logistic - The indicator for Logistic Regression. If set to true, the gradient descent used Sigmoid function as classification algorithm.

Returns

The updated weight matrix

Example

use iron_learn::Tensor;
use iron_learn::gradient_descent::gradient_descent;

let learning_rate: f64 = 0.01;
let w = Tensor::new(vec![2, 1], vec![3.0, 4.0]).unwrap();
let x = Tensor::new(vec![1, 2], vec![3.0, 4.0]).unwrap();
let y = Tensor::new(vec![1, 1], vec![5.0]).unwrap();

let w = gradient_descent(&x, &y, &w, learning_rate, false); // For linear regression