import numpy as np
import matplotlib.pyplot as plt


class RegressionCurve:
    def __init__(self, t0, t1):
        self.theta_0 = t0
        self.theta_1 = t1

    def f(self, x):
        return self.theta_0 + x * self.theta_1

    def E(self, x, y):
        return 0.5 * np.sum((y - self.f(x)) ** 2)

    def standardize(self, x):
        mu = x.mean()
        sigma = x.std()
        return (x - mu) / sigma

    def Regression(self, x, y):
        ETA, diff, count = 1e-3, 1, 0
        error = self.E(x, y)
        while diff > 1e-2:
            self.theta_0 = self.theta_0 - ETA * np.sum(self.f(x) - y)
            self.theta_1 = self.theta_1 - ETA * np.sum((self.f(x) - y) * x)
            current_error = self.E(x, y)
            diff = error - current_error
            error = current_error
            count += 1
            log = 'times:{}  theta_0:{:.3f}  theta_1:{:.3f}  diff:{:.4f}'.format(
                count, self.theta_0, self.theta_1, diff)
            print(log)

    def draw(self, train_x, train_y):
        x = np.linspace(3, -3, 100)
        plt.plot(train_x, train_y, 'o')
        plt.plot(x, self.f(x))
        plt.show()


if __name__=="__main__":
    train = np.loadtxt("click.csv", delimiter=",", skiprows=1)
    train_x, train_y = train[:, 0], train[:, 1]
    theta_0 = np.random.rand()
    theta_1 = np.random.rand()

    regression = RegressionCurve(theta_0, theta_1)
    train_z = regression.standardize(train_x)
    plt.plot(train_z, train_y, 'o')
    plt.show()

    regression.Regression(train_z, train_y)
    regression.draw(train_z, train_y)