案例2：新闻分类(多分类问题)

1. 加载数据集

from keras.datasets import reuters

(train_data, train_labels), (test_data, test_labels) = reuters.load_data(num_words=10000)

将数据限定在10000个最常见出现的单词，8982个训练样本和2264个测试样本

len(train_data)

8982

len(test_data)

2246

train_data[10]

2. 将索引解码为新闻文本

word_index = reuters.get_word_index()
reverse_word_index = dict([(value, key) for (key, value) in word_index.items()])
# Note that our indices were offset by 3
# because 0, 1 and 2 are reserved indices for "padding", "start of sequence", and "unknown".
decoded_newswire = ' '.join([reverse_word_index.get(i - 3, '?') for i in train_data[0]])

train_labels[10]

3. 编码数据

import numpy as np

def vectorize_sequences(sequences, dimension=10000):
    results = np.zeros((len(sequences), dimension))
    for i, sequence in enumerate(sequences):
        results[i, sequence] = 1
    return results

# 将训练数据向量化
x_train = vectorize_sequences(train_data)
# 将测试数据向量化
x_test = vectorize_sequences(test_data)

# 将标签向量化，将标签转化为one-hot
def to_one_hot(labels, dimension=46):
    results = np.zeros((len(labels), dimension))
    for i, label in enumerate(labels):
        results[i, label] = 1
    return results

one_hot_train_labels = to_one_hot(train_labels)
one_hot_test_labels = to_one_hot(test_labels)

from keras.utils.np_utils import to_categorical

one_hot_train_labels = to_categorical(train_labels)
one_hot_test_labels = to_categorical(test_labels)

4. 模型定义

from keras import models
from keras import layers

model = models.Sequential()
model.add(layers.Dense(64, activation='relu', input_shape=(10000,)))
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(46, activation='softmax'))

5. 编译模型

对于这个例子，最好的损失函数是categorical_crossentropy（分类交叉熵），它用于衡量两个概率分布之间的距离

model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])

6. 留出验证集

留出1000个样本作为验证集

x_val = x_train[:1000]
partial_x_train = x_train[1000:]

y_val = one_hot_train_labels[:1000]
partial_y_train = one_hot_train_labels[1000:]

7. 训练模型

history = model.fit(partial_x_train, partial_y_train, epochs=20, batch_size = 512, validation_data = (x_val, y_val))

8. 绘制训练损失和验证损失

import matplotlib.pyplot as plt

loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(1, len(loss) + 1)

plt.plot(epochs, loss, 'bo', label = 'Training loss')
plt.plot(epochs, val_loss, 'b', label = 'Validation loss')
plt.title('Training and validation loss')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()

plt.show()

9. 绘制训练精度和验证精度

plt.clf()     # 清除图像
acc = history.history['acc']
val_acc = history.history['val_acc']

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()

plt.show()

10. 从头开始重新训练一个模型

中间层有64个隐藏神经元

# 从头开始训练一个新的模型
model = models.Sequential()
model.add(layers.Dense(64, activation='relu', input_shape=(10000,)))
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(46, activation='softmax'))

model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(partial_x_train, partial_y_train, epochs=9, batch_size = 512, validation_data = (x_val, y_val))
results = model.evaluate(x_test, one_hot_test_labels)

results

[0.981157986054119, 0.790739091745149]
这种方法可以得到79%的精度

import copy

test_labels_copy = copy.copy(test_labels)
np.random.shuffle(test_labels_copy)
float(np.sum(np.array(test_labels) == np.array(test_labels_copy))) / len(test_labels)

0.19011576135351738 完全随机的精度约为19%

# 在新数据上生成预测结果
predictions = model.predict(x_test)
predictions[0].shape

np.sum(predictions[0])

np.argmax(predictions[0])

11. 处理标签和损失的另一种方法

y_train = np.array(train_labels)
y_test = np.array(test_labels)
model.compile(optimizer='rmsprop', loss='sparse_categorical_crossentropy', metrics=['acc'])

12. 中间层维度足够大的重要性

最终输出是46维的，本代码中间层只有4个隐藏单元，中间层的维度远远小于46

model = models.Sequential()
model.add(layers.Dense(64, activation='relu', input_shape=(10000,)))
model.add(layers.Dense(4, activation='relu'))
model.add(layers.Dense(46, activation='softmax'))

model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(partial_x_train, partial_y_train, epochs=20, batch_size = 128, validation_data = (x_val, y_val))

Epoch 20/20
7982/7982 [==============================] - 2s 274us/step - loss: 0.4369 - acc: 0.8779 - val_loss: 1.7934 - val_acc: 0.7160
验证精度最大约为71%，比前面下降了8%。导致这一下降的主要原因在于，你试图将大量信息（这些信息足够回复46个类别的分割超平面）压缩到维度很小的中间空间

13. 实验

1. 中间层32个

model = models.Sequential()
model.add(layers.Dense(64, activation='relu', input_shape=(10000,)))
model.add(layers.Dense(32, activation='relu'))
model.add(layers.Dense(46, activation='softmax'))

model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(partial_x_train, partial_y_train, epochs=20, batch_size = 128, validation_data = (x_val, y_val))
results = model.evaluate(x_test, one_hot_test_labels)
results

Epoch 20/20
7982/7982 [==============================] - 2s 231us/step - loss: 0.1128 - acc: 0.9564 - val_loss: 1.1904 - val_acc: 0.7970
2246/2246 [==============================] - 0s 157us/step

Out[29]:

[1.4285533854925303, 0.7773820125196835]
精度大约在77%

1. 中间层128个

model = models.Sequential()
model.add(layers.Dense(64, activation='relu', input_shape=(10000,)))
model.add(layers.Dense(128, activation='relu'))
model.add(layers.Dense(46, activation='softmax'))

model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(partial_x_train, partial_y_train, epochs=9, batch_size = 128, validation_data = (x_val, y_val))
results = model.evaluate(x_test, one_hot_test_labels)
results

Epoch 9/9
7982/7982 [==============================] - 2s 237us/step - loss: 0.1593 - acc: 0.9536 - val_loss: 1.0186 - val_acc: 0.8060
2246/2246 [==============================] - 0s 159us/step

Out[31]:

[1.126946303426211, 0.790293855743544]
精度大约在79%
尝试了中间层128个，但是迭代20轮，准确率却只有77%，说明迭代次数过高，出现了过拟合。