[SOLVED] Gradient descent extended function example


Here is my gradient descent code:

import numpy
def gradient_descent(func, grad_func, w_init, n_epochs=100, lr=0.001, verbose=0):
    i = 0
    w = w_init
    while i < n_epochs:
        delta_w = -lr * grad_func(w)
        w = w + delta_w
        if verbose > 0:
            print("f={}; w: {}".format(func(w), w))

        i += 1    
    return w

Now the example that was explained to me is with simple function i.e:
f(p,q) = p^2 + q^2.
The partial derivates are in vector: [2p,2q] and that is clear.
The code for that function and its gradient is:

def f(w):
    return numpy.sum(w*w)

def grad(w):
    return 2*w

So the calculation goes:

# starting point
w_init = numpy.array([10,10])

# learning rate
lr = 0.1

# apply gradient descent
w_opt = gradient_descent(f, grad, w_init, n_epochs=25, lr=lr, verbose=1)

My problem is calculating it for another function such as: f(p,q) = (p^2 + 2q^3). I know the values for partial derivative, but how to implement those values to this particular code?
How to write new function f and gradient function grad for it and to compute it with main function?


Probably you’re looking for something like this:

def f(w):
    return w[0] ** 2 + 2 * w[1] ** 3

def grad(w):
    return np.array([2 * w[0], 3 * w[1] ** 2])

Answered By – jdworzans

Answer Checked By – Terry (BugsFixing Volunteer)

Leave a Reply

Your email address will not be published.