How to use gridsearch to find best learning rate in python?

Question

how can I use gridsearch to find the best learning rate for the gradientDescent?

def computeCost(X,y,theta,theta0):
#TODO
m = len(y)
J = 0
s = 0
#Iterative Solution
for i in range(0,m):
    s = s + ((theta0 + np.dot(theta,X[i])) - y[i])**2
J = s/(2*m)
return J

def gradientDescent(X,y,theta,theta0,alpha,iterations):
#TODO
J_history = np.zeros((iterations,1))
m = len(y)
Y = y.reshape(m,1)
for iter in range(0,iterations):
    s1 = 0
    s0 = 0
    for i in range(0,m):
        s1 += ((theta0 + np.dot(theta,X[i])) - y[i])*X[i]
        s0 += ((theta0 + np.dot(theta,X[i])) - y[i])
    theta0 = theta0 - alpha*s0/m
    theta = theta - alpha*s1/m
    J_history[iter] = computeCost(X,y,theta,theta0)
return theta, J_history

Dataset : https://www.kaggle.com/uciml/pima-indians-diabetes-database

Answer 1

You can use sklearn.model_selection.GridSearchCV() to tune the learning rate. It is possible to define your estimator function(gradientDescent), parameters, and scoring function. See https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html for more information.