AIstudio平台进行深度学习试炼

　　最近百度的paddlepaddl推广力度不是一般的大，在学习paddle同时，也在看其他深度学习框架，懒得搭环境，就借用百度平台进行练手了。由于权限以及各种库版本问题，百度云环境只能用CPU版本的，有点小遗憾。话不多说，直接上代码，keras框架下对电影评论二分类。

#在notebook下编写，首先安装所需的各种库。如果需要进行持久化安装, 需要使用持久化路径

!mkdir /home/aistudio/external-libraries
!pip install keras -t /home/aistudio/external-libraries
!pip install tensorflow -t /home/aistudio/external-libraries

#同时添加如下代码, 这样每次环境(kernel)启动的时候只要运行下方代码即可:

import sys
sys.path.append('/home/aistudio/external-libraries')

# %%writefile './external-libraries/movie.py'
from keras.datasets import imdb
from keras import models
from keras import layers
import numpy as np
import matplotlib.pyplot as plt

(train_data, train_labels), (test_data, test_labels) = imdb.load_data(num_words=10000)
# print('data' ,train_data[0])
# print('label', train_labels[0])
word_index = imdb.get_word_index()
reverse_word_index = dict(
    [(value, key) for (key, value) in word_index.items()])
decoded_review = ''.join(
    [reverse_word_index.get(i - 3, '?') for i in train_data[0]])
print(decoded_review)
def vectorize_sequences(squences, dimension=10000):
    results = np.zeros((len(squences), dimension))
    for i, squence in enumerate(squences):
        results[i, squence] = 1
    # print('results = ', results)
    return results
x_train = vectorize_sequences(train_data)
x_test = vectorize_sequences(test_data)
y_train = np.asarray(train_labels).astype('float32')
y_test = np.asarray(test_labels).astype('float32')

model = models.Sequential()
model.add(layers.Dense(16, activation='relu', input_shape=(10000, )))
model.add(layers.Dense(16, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))
# model.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['accuracy'])
x_val = x_train[:10000]
partial_x_train = x_train[10000:]
# print('partial :', partial_x_train)
y_val = y_train[:10000]
partial_y_train = y_train[10000:]

model.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['acc'])
histroy = model.fit(partial_x_train, partial_y_train, epochs=20, batch_size=512,
                    validation_data=(x_val,y_val))
histroy_dict = histroy.history
# 20组
keys = histroy_dict.keys()
values = histroy_dict.values()
# print('keys:', keys, 'values:', values)
loss_value = histroy_dict['loss']
val_loss_value = histroy_dict['val_loss']
epochs = range(1, len(loss_value) + 1)
plt.plot(epochs, loss_value, 'bo', label='Training loss')
plt.plot(epochs, val_loss_value, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()
plt.clf()
acc = histroy_dict['acc']
val_acc = histroy_dict['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()

下边运行log：

Train on 15000 samples, validate on 10000 samples
Epoch 1/20
15000/15000 [==============================] - 2s 155us/step - loss: 0.5305 - acc: 0.7811 - val_loss: 0.3971 - val_acc: 0.87
Epoch 2/20
15000/15000 [==============================] - 2s 143us/step - loss: 0.3182 - acc: 0.9001 - val_loss: 0.3270 - val_acc: 0.88
Epoch 3/20
15000/15000 [==============================] - 2s 144us/step - loss: 0.2300 - acc: 0.9287 - val_loss: 0.2972 - val_acc: 0.880
Epoch 4/20
15000/15000 [==============================] - 2s 144us/step - loss: 0.1815 - acc: 0.9421 - val_loss: 0.2923 - val_acc: 0.88
Epoch 5/20
15000/15000 [==============================] - 2s 143us/step - loss: 0.1481 - acc: 0.9531 - val_loss: 0.2807 - val_acc: 0.887
Epoch 6/20
15000/15000 [==============================] - 2s 148us/step - loss: 0.1227 - acc: 0.9624 - val_loss: 0.3096 - val_acc: 0.878
Epoch 7/20
15000/15000 [==============================] - 2s 143us/step - loss: 0.1021 - acc: 0.9709 - val_loss: 0.3169 - val_acc: 0.87
Epoch 8/20
15000/15000 [==============================] - 2s 144us/step - loss: 0.0872 - acc: 0.9751 - val_loss: 0.3257 - val_acc: 0.88
Epoch 9/20
15000/15000 [==============================] - 2s 144us/step - loss: 0.0724 - acc: 0.9814 - val_loss: 0.3503 - val_acc: 0.882
Epoch 10/20
15000/15000 [==============================] - 2s 143us/step - loss: 0.0610 - acc: 0.9846 - val_loss: 0.3702 - val_acc: 0.88
Epoch 11/20
15000/15000 [==============================] - 2s 143us/step - loss: 0.0489 - acc: 0.9901 - val_loss: 0.4122 - val_acc: 0.878
Epoch 12/20
15000/15000 [==============================] - 2s 144us/step - loss: 0.0417 - acc: 0.9915 - val_loss: 0.4241 - val_acc: 0.876
Epoch 13/20
15000/15000 [==============================] - 2s 147us/step - loss: 0.0327 - acc: 0.9947 - val_loss: 0.4542 - val_acc: 0.87
Epoch 14/20
15000/15000 [==============================] - 2s 145us/step - loss: 0.0272 - acc: 0.9951 - val_loss: 0.4983 - val_acc: 0.86
Epoch 15/20
15000/15000 [==============================] - 2s 147us/step - loss: 0.0213 - acc: 0.9965 - val_loss: 0.5144 - val_acc: 0.86
Epoch 16/20
15000/15000 [==============================] - 2s 147us/step - loss: 0.0171 - acc: 0.9975 - val_loss: 0.5459 - val_acc: 0.870
Epoch 17/20
15000/15000 [==============================] - 2s 152us/step - loss: 0.0123 - acc: 0.9985 - val_loss: 0.5868 - val_acc: 0.868
Epoch 18/20
15000/15000 [==============================] - 2s 146us/step - loss: 0.0146 - acc: 0.9967 - val_loss: 0.6150 - val_acc: 0.867
Epoch 19/20
15000/15000 [==============================] - 2s 145us/step - loss: 0.0056 - acc: 0.9996 - val_loss: 0.6424 - val_acc: 0.865
Epoch 20/20
15000/15000 [==============================] - 2s 150us/step - loss: 0.0089 - acc: 0.9979 - val_loss: 0.6818 - val_acc: 0.865

训练与验证的损失和精度图：

这里需要注意，由于用的是百度的服务器，下载的数据集等内容重启就会清除，为了永久保存，需要修改/keras/utils/data_utils.py文件中的路径信息，如下

176行的
cache_dir = os.path.join(os.path.expanduser('~'), '.keras')
修改为：
cache_dir = os.path.join(os.path.expanduser('~'), 'external-libraries/keras/')

这样可以确保数据集下载一次，后续不用再下载，当然，如果是自己的主机，可以不用做该步操作。

代码在《Python深度学习》第三章