import csv
import numpy as np
import matplotlib.pyplot as plt
from random import uniform


def J(X, y, theta):
    preds = np.squeeze(np.matmul(X, theta))
    temp =  preds - np.squeeze(y)
    return np.sqrt(np.sum(np.matmul(np.transpose(temp), temp)))

#########################################################################
#   Read Data
#########################################################################


def load_dataset(pat):
    with open(pat) as csv_file:
        csv_reader = csv.reader(csv_file, delimiter=',')
        next(csv_reader) # to skip the header file
        X = []  # inputs
        y = []  # outputs
        for row in csv_reader:
            X.append([float(row[0]), 1])
            y.append(float(row[1]))

    X = np.asarray(X)
    y = np.asarray(y)

    return X, y

#########################################################################
#   Gradient Descent method
#########################################################################

def gradient_descent(theta, lr, tol, X, y):

    it = 0
    Js = []
    while True:
        t_0 = 0
        t_1 = 0
        
        for j in range(len(y)):
            t_0 +=  (theta[0] * X[j][0] + theta[1] - y[j]) * X[j][0]
            t_1 +=  theta[0] * X[j][0] + theta[1] - y[j]

        t_0 = t_0 / len(y)
        t_1 = t_0 / len(y)

        theta_old = np.copy(theta)
    
        theta[0] = theta[0] - lr * t_0
        theta[1] = theta[1] - lr * t_1
        
        it += 1
        
        delta = np.sum(np.abs(theta - theta_old))
        
        Js.append(J(X,y, theta_gd))
        
        print('[%d] Solution (Gradient descent): J=%.5f, theta=(%.2f, %.2f), delta=%.5f' % (it, Js[-1], theta_gd[0], theta_gd[1], delta))
        
        
        plt.ion()
        plt.figure(1)
        plt.plot(range(len(Js)), Js, '-ko')
        plt.grid()
        plt.xlabel('Iterations')
        plt.ylabel('J()')
        plt.show()
        plt.pause(0.1)

                
        if  delta < tol:
            break
        
    return theta, Js


learning_rate = 0.001
tolerance = 0.0001
theta_gd = np.asarray([uniform(0., 0.5), uniform(750., 1000.)])

Features, outs = load_dataset('/Users/hugomcp/Desktop/pizza.csv')

theta_gd, evolution_J = gradient_descent(theta_gd, learning_rate, tolerance, Features, outs)


input('Close app?')