import tensorflow as tf
import matplotlib.pylab as plt
import numpy as np

# 呼叫資料
boston_house = tf.keras.datasets.boston_housing
(train_x, train_y), (test_x, test_y) = boston_house.load_data(test_split=0.1)
print(train_x.shape,train_y.shape)

#   樣本資料歸一化
    #   測試樣本
X = (train_x - train_x.min(axis = 0)) / (train_x.max(axis = 0) - train_x.min(axis = 0))
    
 
#   訓練樣本
t_X = (test_x - test_x.min(axis = 0)) / (test_x.max(axis = 0) - test_x.min(axis = 0))
print('X:',X,X.shape)

Y = train_y.reshape(-1,1)                    #   將房價y轉化為列向量
    #   測試集
t_Y = test_y.reshape(-1,1)
print('Y.shape:',Y.shape)
#   超引數
iter =20000              #   迭代次數
learn_rate = 0.01      #   學習率
 

#   設定模型引數初始值
w = np.random.randn(13,1)
W = tf.Variable(w)
loss_list = []
test_loss_list = []
# print('W:{},W.shape:{}'.format(W,W.shape))
for i in range(iter):
    with tf.GradientTape() as tape:
        pre_price = tf.matmul(X,W)
        pre_test = tf.matmul(t_X, W)
        loss = 0.5 * (tf.reduce_mean(pow(Y - pre_price,2)))
        test_loss 
 
= 0.5 * (tf.reduce_mean(pow(t_Y - pre_test,2)))
    loss_list.append(loss)
    test_loss_list.append(test_loss)
    dloss_dw = tape.gradient(loss,W)                        #   偏導也是一個2維向量
    # print('dloss_dw:',dloss_dw)
    W.assign_sub(learn_rate * dloss_dw)
    if i % 1000 == 0:
        print('訓練集：i:{},loss:{},W:{}'.format(i,loss,W))
        print('測試集：i:{},test_loss:{},W:{}'.format(i, test_loss, W))
pre_test = tf.matmul(t_X,W)
test_loss = 0.5 * (tf.reduce_mean(pow(t_Y - pre_test,2)))
print('測試集loss為:',test_loss)
#   畫圖
plt.rcParams["font.family"] = 'SimHei'              # 將字型改為中文
plt.rcParams['axes.unicode_minus'] = False          # 設定了中文字型預設後，座標的"-"號無法顯示，設定這個引數就可以避免

plt.subplot(221)
plt.plot(loss_list,label = '訓練集損失值')
plt.legend()
plt.subplot(223)
plt.plot(test_loss_list,label = '測試集損失值')
plt.legend()
plt.show()