第三章 神经网络入门

下面展示一些 内联代码片。

""" 
预测房价--回归问题  
"""
# 1.价值波士顿房价数据
from keras.datasets import boston_housing
(train_data, train_targets), (test_data, test_targets) = boston_housing.load_data()     # 切分数据
print(train_data.shape)              # 2D连续型数据(404, 13)
print(test_data.shape)
print(train_data.shape[1])           # 特征数量13
print(train_targets.shape)

# 2.准备数据：数据标准化(对特征)
mean = train_data.mean(axis=0)   # 平均值  0代表列的计算，1代表行的计算
train_data -= mean
std = train_data.std(axis=0)     # 标准差
train_data /= std                # 标准化

test_data -=mean            # 用于测试数据标准化的均值和标准差都是在训练数据上计算到的 
test_data /=std


# 3.构建网络：定义模型
from keras import models
from keras import layers

def build_model():
    model = models.Sequential()             # 因为需要将同一个模型多次实例化，所以用一个函数来构建模型
    model.add(layers.Dense(64, activation='relu', input_shape=(train_data.shape[1],)))
    model.add(layers.Dense(64, activation='relu'))
    model.add(layers.Dense(1))              # 无激活函数，标量回归的典型设置
    
    model.compile(optimizer='rmsprop',      # 优化器rmsprop
                  loss='mse',               # mse损失函数（均方误差）MSE
                  metrics=['mae'])          # 指标：平均绝对误差MAE
    return model
    model.summary()


# 4.K折交叉验证
import numpy as np

k = 4
num_val_samples = len(train_data) // k      # 浮点数   num_val_samples是1/4的数据长度
num_epochs = 100
all_scores = []            # 记录平均绝对误差MAE

for i in range(k):
    print('processing fold #', i)           # 准备验证数据，第k个分区的数据
    val_data = train_data[i * num_val_samples : (i+1) * num_val_samples]         # 验证数据
    val_targets = train_targets[i * num_val_samples : (i+1) * num_val_samples]   # 验证标签
    
    partial_train_data = np.concatenate(                   # 准备训练数据：其他所有分区的数据 将数据连接起来
        [train_data[ : i * num_val_samples],
         train_data[(i+1) * num_val_samples : ]],
        axis=0)
    partial_train_targets = np.concatenate(                # 准备训练标签：其他所有分区的标签
        [train_targets[ : i * num_val_samples],
         train_targets[(i+1) * num_val_samples : ]],
        axis=0)
    
    model = build_model()         # 构建Keras模型（已编译）
    model.fit(partial_train_data, 
              partial_train_targets,
              epochs = num_epochs, 
              batch_size = 1, 
              verbose = 0)          # 训练模型（静默模式，verbose=0）
    
    # model.evaluate：在验证数据上评估模型
    val_mse, val_mae = model.evaluate(val_data, val_targets, verbose=0)
    all_scores.append(val_mae)                       # 记录平均绝对误差MAE
    print('第%i次平均绝对误差MAE:', all_scores[i]) 
    
print('4次平均绝对误差MAE的平均值:', np.mean(all_scores))     # 4次平均绝对误差MAE的平均值


# 5.保存每折的验证结果
num_epochs = 500
all_mae_histories = []      # 指标：平均绝对误差MAE
for i in range(k):
    print('processing fold #', i)
    val_data = train_data[i * num_val_samples : (i+1) * num_val_samples]         # 验证数据
    val_targets = train_targets[i * num_val_samples : (i+1) * num_val_samples]   # 验证标签
    
    partial_train_data = np.concatenate(                   # 准备训练数据：其他所有分区的数据
        [train_data[ : i * num_val_samples],
         train_data[(i+1) * num_val_samples : ]],
        axis=0)
    partial_train_targets = np.concatenate(                # 准备训练标签：其他所有分区的标签
        [train_targets[ : i * num_val_samples],
         train_targets[(i+1) * num_val_samples : ]],
        axis=0)
    
    model = build_model()
    History = model.fit(partial_train_data, 
                        partial_train_targets,
                        validation_data = (val_data, val_targets),    # 将验证集传入validation_data()来完成验证
                        epochs = num_epochs, 
                        batch_size = 1, 
                        verbose = 0)
    
    mae_history = History.history['val_mae']         # 记录平均绝对误差MAE(val_mean_absolute_error)
    # mae_history = History.history['val_mean_absolute_error']       # 错误：KeyError: 'val_mean_absolute_error'
    all_mae_histories.append(mae_history)
    
print('500次平均绝对误差MAE:', np.mean(all_mae_histories))     # 500次平均绝对误差MAE    
print(History.history.keys())          # dict_keys(['val_loss', 'val_mae', 'loss', 'mae'])


# 6.计算所有轮次中的K折验证分数平均值
average_mae_history = [np.mean([x[i] for x in all_mae_histories]) for i in range(num_epochs)]


# 7.绘制验证分数
import matplotlib.pyplot as plt

plt.plot(range(1, len(average_mae_history) + 1), average_mae_history)
plt.xlabel('Epochs')
plt.ylabel('Validation MAE')
plt.show()
 
# 8.绘制验证分数(删除前10个数据点)
def smooth_curve(points, factor=0.9):
    smoothed_points = []
    for point in points:
        if smoothed_points:
            previous = smoothed_points[-1]
            smoothed_points.append(previous * factor + point * (1 - factor))
        else:
            smoothed_points.append(point)
    return smoothed_points               # 指数移动平均值

smooth_mae_history = smooth_curve(average_mae_history[10:])    # 500删除前10

plt.plot(range(1, len(smooth_mae_history) + 1), smooth_mae_history)
plt.xlabel('Epochs')
plt.ylabel('Validation MAE')
plt.show()


# 9.训练最终模型
model = build_model()
model.fit(train_data, 
          train_targets, 
          epochs=80, 
          batch_size=16, 
          verbose=0)

# model.evaluate：在验证数据上评估模型
test_mse_score, test_mae_score = model.evaluate(test_data, test_targets)  
 
print('（均方误差）MSE:',test_mse_score)           # mse损失函数（均方误差）MSE
print('平均绝对误差MAE', test_mae_score)           # 指标：平均绝对误差MAE

代码运行结果：无