3.1.3全数据梯度下降法

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

for_mat = 'Advertising.csv'
advertising = pd.read_csv(for_mat)
advertising.head()

"""
以TV为自变量，以sales为因变量
对TV进行标准化，对sales进行中心化
"""
TV = advertising["TV"].values
sales = advertising["sales"].values
scaled_TV = (TV - np.mean(TV)) / np.std(TV)
centered_sales = sales - np.mean(sales)
n = len(scaled_TV)
"""
画出RSS随着w变化的曲线
画出RSS在w=0处的切线；画出RSS的最小值点
用函数plot_rss_w实现
"""


def plot_rss_w():
    w_vec = np.linspace(-2, 10, 100)  # 权重向量
    rss_vec = []  # 记录不同权重下的RSS值
    for w in w_vec:
        rss = np.sum((centered_sales - scaled_TV * w) ** 2) / 2 / n
        rss_vec.append(rss)

    w_0 = 0
    rss_0 = np.sum((centered_sales - scaled_TV * w_0) ** 2) / 2 / n
    w_min = np.sum(scaled_TV * centered_sales) / np.sum(scaled_TV ** 2)
    rss_min = np.sum((centered_sales - scaled_TV * w_min) ** 2) / 2 / n
    plt.plot(w_vec, rss_vec)
    plt.scatter(w_0, rss_0, s=100, c="y", marker="o")
    plt.scatter(w_min, rss_min, s=100, c="r", marker="D")
    plt.plot(np.linspace(-1, 1, 50),
             np.linspace(-1, 1, 50) * np.mean(-scaled_TV * centered_sales) + np.sum(centered_sales ** 2) / 2 / n, '--',
             linewidth=2.0)
    plt.xlabel("w", fontsize=16)
    plt.ylabel("RSS", fontsize=16)
    plt.show()


# plot_rss_w()

"""
使用全数据梯度下降法迭代更新w
"""

w, lr = 0, 0.1
input_i, target = scaled_TV, centered_sales
w_record = []
loss_record = []
for iterm in range(20):
    pred = input_i * w
    loss = np.sum((pred - target) ** 2) / 2 / n

    w_record.append(w)
    loss_record.append(loss)

    delta = pred - target
    w = w - lr - np.sum(input_i * delta) / n

    if iterm % 5 == 0 or iterm == 19:
        print("iterm: %3d;Loss: %0.3f" % (iterm, loss))
        w_record.append(w)
        loss_record.append(np.sum((pred - target)**2)/2/n)