gistfile1.py

def stochastic_linear_gradient_descent(X, y, theta, alpha, tolerance):
    converged = False
    prev_cost = 0
    size = X.shape[0]
    XT = X.transpose()
    while not converged:
        for i in range (size):
            sample_x = X[i,:]
            hypothesis = np.dot(sample_x, theta)
            loss = hypothesis - y
            curr_cost = compute_lin_cost(theta, X, y)
            grad = (1.0/size) * np.dot(XT, loss)
            theta -= alpha * grad
            if (abs(curr_cost - prev_cost) < tolerance):
                converged = True
                break
            else:
                converged = False
            prev_cost = curr_cost
            print("cost " , curr_cost)
    return theta