深度学习-多分类问题

#导入数据
from keras.datasets import reuters
(train_data, train_labels), (test_data, test_labels) = reuters.load_data(
num_words=10000)# 数num_words=10000 将数据限定为前10 000 个最常出现的单词

Using TensorFlow backend.


Downloading data from https://s3.amazonaws.com/text-datasets/reuters.npz
2113536/2110848 [==============================] - 7s 4us/step

len(train_data)# 训练数据为8982
len(test_data)# 测试数据为2246
# 训练集为多个list的组合
#测试集为一列数字

2246

#对索引解码
word_index = reuters.get_word_index()
# 返回的结果为字典
reverse_word_index = dict([(value, key) for (key, value) in word_index.items()])
decoded_newswire = ' '.join([reverse_word_index.get(i - 3, '?') for i in
train_data[0]])
# 通过字典可以讲字符组合

Downloading data from https://s3.amazonaws.com/text-datasets/reuters_word_index.json
557056/550378 [==============================] - 5s 10us/step

# 编码数据向量化
import numpy as np
def vectorize_sequences(sequences, dimension=10000):
    results = np.zeros((len(sequences), dimension))# 0矩阵,2262行1000列
    for i, sequence in enumerate(sequences):#enumerate() 函数用于将一个可遍历的数据对象组合为一个索引序列，同时列出数据和数据下标
        results[i, sequence] = 1.
    return results
x_train = vectorize_sequences(train_data)
x_test = vectorize_sequences(test_data)

# 标签向量化
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)

# 定义模型
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'))# 增加节点
model.add(layers.Dense(46, activation='softmax'))# 最终的分类为46类

# 添加模型参数
model.compile(optimizer='rmsprop',# 模型优化器
    loss='categorical_crossentropy',#损失函数为分类交叉熵
    metrics=['accuracy'])# 目标函数为准确率

# 设置验证集
x_val = x_train[:1000]# 验证1000个样本
partial_x_train = x_train[1000:]#剩下的训练
y_val = one_hot_train_labels[:1000]
partial_y_train = one_hot_train_labels[1000:]

# 开始训练
history = model.fit(partial_x_train,
                    partial_y_train,
                    epochs=20,# 执行20次迭代
                    batch_size=512,
                    validation_data=(x_val, y_val))# 验证

Train on 7982 samples, validate on 1000 samples
Epoch 1/20
7982/7982 [==============================] - 1s 98us/step - loss: 2.7797 - accuracy: 0.5312 - val_loss: 1.8418 - val_accuracy: 0.6360
Epoch 2/20
7982/7982 [==============================] - 0s 62us/step - loss: 1.4911 - accuracy: 0.6974 - val_loss: 1.3655 - val_accuracy: 0.7100
Epoch 3/20
7982/7982 [==============================] - 0s 60us/step - loss: 1.0956 - accuracy: 0.7602 - val_loss: 1.1734 - val_accuracy: 0.7400
Epoch 4/20
7982/7982 [==============================] - 0s 59us/step - loss: 0.8541 - accuracy: 0.8187 - val_loss: 1.0490 - val_accuracy: 0.7850
Epoch 5/20
7982/7982 [==============================] - 0s 59us/step - loss: 0.6788 - accuracy: 0.8568 - val_loss: 0.9739 - val_accuracy: 0.7950
Epoch 6/20
7982/7982 [==============================] - 0s 58us/step - loss: 0.5393 - accuracy: 0.8894 - val_loss: 0.9662 - val_accuracy: 0.7860
Epoch 7/20
7982/7982 [==============================] - 0s 59us/step - loss: 0.4347 - accuracy: 0.9090 - val_loss: 0.9183 - val_accuracy: 0.8060
Epoch 8/20
7982/7982 [==============================] - 0s 58us/step - loss: 0.3533 - accuracy: 0.9270 - val_loss: 0.9319 - val_accuracy: 0.7990
Epoch 9/20
7982/7982 [==============================] - 0s 61us/step - loss: 0.2924 - accuracy: 0.9371 - val_loss: 0.8917 - val_accuracy: 0.8180
Epoch 10/20
7982/7982 [==============================] - 0s 58us/step - loss: 0.2459 - accuracy: 0.9444 - val_loss: 0.9028 - val_accuracy: 0.8180
Epoch 11/20
7982/7982 [==============================] - 0s 58us/step - loss: 0.2146 - accuracy: 0.9465 - val_loss: 0.9152 - val_accuracy: 0.8160
Epoch 12/20
7982/7982 [==============================] - 0s 58us/step - loss: 0.1849 - accuracy: 0.9509 - val_loss: 0.9823 - val_accuracy: 0.8040
Epoch 13/20
7982/7982 [==============================] - 0s 60us/step - loss: 0.1668 - accuracy: 0.9530 - val_loss: 0.9608 - val_accuracy: 0.8070
Epoch 14/20
7982/7982 [==============================] - 0s 62us/step - loss: 0.1562 - accuracy: 0.9545 - val_loss: 0.9910 - val_accuracy: 0.8080
Epoch 15/20
7982/7982 [==============================] - 0s 61us/step - loss: 0.1412 - accuracy: 0.9541 - val_loss: 0.9723 - val_accuracy: 0.8130
Epoch 16/20
7982/7982 [==============================] - 0s 60us/step - loss: 0.1300 - accuracy: 0.9573 - val_loss: 1.0508 - val_accuracy: 0.8030
Epoch 17/20
7982/7982 [==============================] - 0s 58us/step - loss: 0.1265 - accuracy: 0.9565 - val_loss: 1.0783 - val_accuracy: 0.7950
Epoch 18/20
7982/7982 [==============================] - 0s 59us/step - loss: 0.1201 - accuracy: 0.9546 - val_loss: 1.0738 - val_accuracy: 0.8030
Epoch 19/20
7982/7982 [==============================] - 0s 59us/step - loss: 0.1163 - accuracy: 0.9554 - val_loss: 1.0397 - val_accuracy: 0.8160
Epoch 20/20
7982/7982 [==============================] - 0s 58us/step - loss: 0.1122 - accuracy: 0.9568 - val_loss: 1.0825 - val_accuracy: 0.8090

import matplotlib.pyplot as plt# 导入matplotlib

loss = history.history['loss']# 提取损失值
val_loss = history.history['val_loss']# 验证损失
epochs = range(1, len(loss) + 1)# 迭代次数
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()

# 绘制精确度
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
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()

# 重新训练神经网络
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,# 这里的迭代次数选择9
            batch_size=512,
            validation_data=(x_val, y_val))
results = model.evaluate(x_test, one_hot_test_labels)
results

Train on 7982 samples, validate on 1000 samples
Epoch 1/9
7982/7982 [==============================] - 1s 82us/step - loss: 2.6067 - accuracy: 0.5381 - val_loss: 1.7146 - val_accuracy: 0.6550
Epoch 2/9
7982/7982 [==============================] - 0s 58us/step - loss: 1.3859 - accuracy: 0.7076 - val_loss: 1.2856 - val_accuracy: 0.7240
Epoch 3/9
7982/7982 [==============================] - 0s 58us/step - loss: 1.0247 - accuracy: 0.7813 - val_loss: 1.1473 - val_accuracy: 0.7420
Epoch 4/9
7982/7982 [==============================] - 0s 58us/step - loss: 0.8143 - accuracy: 0.8235 - val_loss: 1.0376 - val_accuracy: 0.7820
Epoch 5/9
7982/7982 [==============================] - 0s 58us/step - loss: 0.6458 - accuracy: 0.8648 - val_loss: 0.9959 - val_accuracy: 0.7920
Epoch 6/9
7982/7982 [==============================] - 0s 57us/step - loss: 0.5195 - accuracy: 0.8901 - val_loss: 0.9377 - val_accuracy: 0.8040
Epoch 7/9
7982/7982 [==============================] - 0s 56us/step - loss: 0.4196 - accuracy: 0.9128 - val_loss: 0.9240 - val_accuracy: 0.8120
Epoch 8/9
7982/7982 [==============================] - 0s 55us/step - loss: 0.3369 - accuracy: 0.9295 - val_loss: 0.8785 - val_accuracy: 0.8150
Epoch 9/9
7982/7982 [==============================] - 0s 55us/step - loss: 0.2789 - accuracy: 0.9376 - val_loss: 0.9101 - val_accuracy: 0.8060
2246/2246 [==============================] - 0s 69us/step





[0.9737234661224583, 0.7867319583892822]

# 对新数据进行预测
predictions = model.predict(x_test)
predictions[0].shape
# predictions为46个类别的预测概率
# 判定的依据是最大概率为可能的分类

(46,)