import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, datasets, losses
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

plt.rcParams['font.sans-serif'] = ['SimHei']  # 用來正常顯示中文標簽
plt.rcParams['axes.unicode_minus'] = False  # 用來正常顯示負號

import matplotlib
print(matplotlib.matplotlib_fname())

# 在線下載汽車效能數據集
dataset_path = keras.utils.get_file("auto-mpg.data", 
"http://archive.ics.uci.edu/ml/machine-learning-databases/auto-mpg/auto-mpg.data")
# 字段有效能（公里數每加侖） 氣缸數 排量 馬力 重量 加速度 型號年份 產地
column_names = ['MPG','Cylinders','Displacement','Horsepower','Weight',
 'Acceleration', 'Model Year', 'Origin']
raw_dataset = pd.read_csv(dataset_path, names=column_names,
 na_values = "?", comment='\t',
 sep=" ", skipinitialspace=True)

dataset = raw_dataset.copy()
# 查看部分數據
dataset.head(8)

begin = dataset.isna().sum()  # 統(tǒng)計空白數據
dataset = dataset.dropna()  # 刪除空白數據項
end = dataset.isna().sum()  # 再次統(tǒng)計空白數據
print(f'========統(tǒng)計前:========\n{begin}\n========統(tǒng)計后:========\n{end}')
# 輸出結果:
========統(tǒng)計前:========
MPG             0
Cylinders       0
Displacement    0
Horsepower      6
Weight          0
Acceleration    0
Model Year      0
Origin          0
dtype: int64
========統(tǒng)計后:========
MPG             0
Cylinders       0
Displacement    0
Horsepower      0
Weight          0
Acceleration    0
Model Year      0
Origin          0
dtype: int64

# 先彈出（刪除并返回）Origin這一列
origin = dataset.pop('Origin')
# 根據origin列的數據來寫入新的3個列
dataset['USA'] = (origin == 1) * 1.0
dataset['Europe'] = (origin == 2) * 1.0
dataset['Japan'] = (origin == 3) * 1.0
dataset.tail()

# 切分為訓練集和測試集
train_dataset = dataset.sample(frac=0.8, random_state=0)
test_dataset = dataset.drop(train_dataset.index)

# 查看訓練集的輸入x的統(tǒng)計數據
train_stats = train_dataset.describe()
train_stats = train_stats.transpose()  # 轉置
train_stats

# 移動MPG油耗效能這一列為真實標簽Y
train_labels = train_dataset.pop('MPG')
test_labels = test_dataset.pop('MPG')

# 標準化數據
def norm(x):
    # 減去每個字段的均值并除以標準差
    return (x - train_stats['mean']) / train_stats['std']
 
normed_train_data = norm(train_dataset)  # 標準化訓練集
normed_test_data = norm(test_dataset)  # 標準化測試集
print(f'訓練集大小:{normed_train_data.shape}  {train_labels.shape}')
print(f"測試集大小:{normed_test_data.shape}  {test_labels.shape}")
normed_train_data  # 查看數據處理后的訓練集
# 利用切分的訓練集構建數據集對象
train_db = tf.data.Dataset.from_tensor_slices((normed_train_data.values, train_labels.values))  # 構建Dataset對象
# 為防止數據標準化后出現(xiàn)梯度彌散，切分訓練集，分為幾個batch，加速計算
train_db = train_db.shuffle(100).batch(32)  # 隨機打散，批量化
 
# 輸出結果:
訓練集大小:(314, 9)  (314,)
測試集大小:(78, 9)  (78,)

# 創(chuàng)建同y軸的上下三聯(lián)圖，需要plt.subplots()
fig, axes = plt.subplots(nrows=1,ncols=3,sharey=True,figsize=(20,6)) 
fig.suptitle('各個字段對MPG的影響', fontsize=16)
#subplot1
axes[0].scatter(train_dataset['Cylinders'].to_numpy(), train_labels.to_numpy(), color='b',s=20)
axes[0].set_xlabel('Cylinders', fontsize=13)
axes[0].set_ylabel('MPG', fontsize=14)
#subplot2
axes[1].scatter(train_dataset['Displacement'].to_numpy(), train_labels.to_numpy(), color='b',s=20)
axes[1].set_xlabel('Displacement', fontsize=13)
#subplot3
axes[2].scatter(train_dataset['Weight'].to_numpy(), train_labels.to_numpy(), color='b',s=20)
axes[2].set_xlabel('Weight', fontsize=13)
 
#展示圖片
plt.show()

class Network(keras.Model):
    # 回歸網絡模型
    def __init__(self):
        super(Network, self).__init__()
        # 創(chuàng)建3個全連接層
        self.fc1 = layers.Dense(64, activation='relu', name='Layer1')
        self.fc2 = layers.Dense(64, activation='relu', name='Layer2')
        self.fc3 = layers.Dense(1, name='OutputLayer')
        
    def call(self, inputs, training=None, mask=None):
        # 依次通過3個全連接層
        x = self.fc1(inputs)
        x = self.fc2(x)
        x = self.fc3(x)
        return x

# 創(chuàng)建網絡類實例
model = Network()
# 通過build函數完成內部張量的創(chuàng)建（其中4為任意設置的batch量，9為輸入特征長度）
model.build(input_shape=(4, 9))
# 打印網絡信息
model.summary()
# 創(chuàng)建優(yōu)化器，指定學習率為0.001
optimizers = tf.keras.optimizers.RMSprop(0.001)

# 接下來實現(xiàn)網絡訓練部分。通過 Epoch 和 Step 組成的雙層循環(huán)訓練網絡，共訓練 200 個 Epoch
Epoch =  np.arange(0, 200)
train_MAE = np.zeros(200)
test_MAE = np.zeros(200)
 
for epoch in range(200):
    for step, (x, y) in enumerate(train_db):  # 遍歷依次訓練集
        # 梯度記錄器
        with tf.GradientTape() as tape:
            out = model(x)  # 通過網絡獲得輸出
            loss = tf.reduce_mean(losses.MSE(y, out))
            mae_loss = tf.reduce_mean(losses.MAE(y, out)) # 計算 MAE
            # 計算梯度并更新
            grads = tape.gradient(loss, model.trainable_variables)
            Optimizers.apply_gradients(zip(grads, model.trainable_variables))
    train_MAE[epoch] = float(mae_loss)
    out = model(tf.constant(normed_test_data.values))
    test_MAE[epoch] = tf.reduce_mean(losses.MAE(test_labels, out))

plt.xlabel('Epoch')
plt.ylabel('MAE')
plt.plot(Epoch, train_MAE, label="Train")
plt.plot(Epoch, test_MAE, label="Test")
plt.title('汽車油耗實戰(zhàn)')
plt.legend(['Train', 'Test'])  # 設置折線名稱
plt.show()