Tags: DeepLearning Tool
Keras介绍
keras是一个深度学习的high level API, 由python实现，支持Tesorflow, Theano, CNTK作为后端(也称为backend), 最近keras也成为Tensorflow的官方high level API, 因此和Tensorflow的适配性更好。keras支持简洁快速的原型设计, 支持CNN和RNN, 无缝CPU/GPU切换。此外keras模型也能直接转换为CoreML模型用在IOS设备上。
如果你熟悉深度学习基础概念, Keras很容易上手进行快速复现,因为不需要自己实现很多layer, 门槛比较低。 适合用于快速训练模型部署到生产环境,研究的话还是Tensorflow和caffe2+Pytorch吧,但是Tensorflow感觉实在是太复杂了, 接口太冗余。
Sequential Model
序列模型是多个网络层的线性堆叠, 可以传入layer list或者调用add()方法将layer加入模型。
from keras.models import Sequential
from keras.layers import Dense, Activation
# 传list
model1 = Sequential([Dense(32, input_dim=784), Activation('relu'), Dense(10), Activation('softmax')])
# add() 方法
model2 = Sequential()
model2.add(Dense(32, input_shape=(784,)))
model2.add(Activation('relu'))
model2.add(Dense(10))
model2.add(Activation('softmax'))
定义好模型之后需要调用compile方法对模型进行配置, 此时传入三个参数: optimizer, loss, metrics(如accuracy), loss和metrics都可以自定义。
# For a multi-class classification problem
model.compile(optimizer='rmsprop',
loss='categorical_crossentropy',
metrics=['accuracy'])
# For a binary classification problem
model.compile(optimizer='rmsprop',
loss='binary_crossentropy',
metrics=['accuracy'])
# For a mean squared error regression problem
model.compile(optimizer='rmsprop',
loss='mse')
# For custom metrics
import keras.backend as K
def mean_pred(y_true, y_pred):
return K.mean(y_pred)
model.compile(optimizer='rmsprop',
loss='binary_crossentropy',
metrics=['accuracy', mean_pred])
定义好模型和配置之后就可以使用fit()和fit_generator()方法传入进行训练。
model.fit(x_train, y_train,
epochs=20,
batch_size=128)
训练完了之后就可以调用evaluate()方法对训练好的模型进行评估。
score = model.evaluate(x_test, y_test, batch_size=128)
感觉真的是很简单:) user-friendly, made for human, 不像Tensorflow...
下面看一个Sequential Model的Mnist手写字符分类(深度学习hello world)例子吧。
keras内部有一些数据集接口, 包括 CIFAR10图片分类数据,CIFAR-100图片数据, IMDB电影评论分类, 路透社新闻主题分类, MNIST手写字体识别, Fashion-MNIST, 波士顿房价数据(回归模型)。可以用来练习或者测试自己的model。
感知机模型
from keras import layers
from keras import models
from keras.datasets import mnist
from keras.utils import to_categorical # convert int labels to one-hot vector
# define model
model = models.Sequential()
model.add(layers.Dense(128, activation='relu', input_dim=784))
model.add(layers.Dropout(0.5))
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dropout(0.5))
model.add(layers.Dense(10, activation='softmax'))
# print the model
model.summary
# load data
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()
train_images = train_images.astype('float32')/255 # normalize to 0~1
test_images = test_images.astype('float32')/255
train_images = train_images.reshape((60000,-1))
test_images = test_images.reshape((10000,-1))
# convert to one-hot vectors
train_labels = to_categorical(train_labels)
test_labels = to_categorical(test_labels)
# define training config
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
# train the model
model.fit(train_images, train_labels, epochs=5, batch_size=64)
# evaluate the model
test_loss, test_accuracy = model.evaluate(test_images, test_labels)
print("test loss:", test_loss)
print("test accuracy:", test_accuracy)
# 输出
60000/60000 [==============================] - 2s 37us/step - loss: 0.6188 - acc: 0.8094
Epoch 2/5
60000/60000 [==============================] - 2s 32us/step - loss: 0.3359 - acc: 0.9093
Epoch 3/5
60000/60000 [==============================] - 2s 32us/step - loss: 0.2908 - acc: 0.9231
Epoch 4/5
60000/60000 [==============================] - 2s 32us/step - loss: 0.2699 - acc: 0.9317
Epoch 5/5
60000/60000 [==============================] - 2s 32us/step - loss: 0.2650 - acc: 0.9347
10000/10000 [==============================] - 0s 23us/step
test loss: 0.16157680716912728
test accuracy: 0.9622
卷积模型
from keras.datasets import mnist
from keras.utils import to_categorical # convert int labels to one-hot vector
# define model
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10, activation='softmax'))
# print the model
print(model.summary)
# load data
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()
train_images = train_images.reshape((60000, 28, 28, 1))
train_images = train_images.astype('float32')/255 # normalize to 0~1
test_images = test_images.reshape((10000, 28, 28, 1))
test_images = test_images.astype('float32')/255
# convert to one-hot vectors
train_labels = to_categorical(train_labels)
test_labels = to_categorical(test_labels)
# define training config
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
# train the model
model.fit(train_images, train_labels, epochs=5, batch_size=64)
# evaluate the model
test_loss, test_accuracy = model.evaluate(test_images, test_labels)
print("test loss:", test_loss)
print("test accuracy:", test_accuracy)
# 输出
60000/60000 [==============================] - 34s 565us/step - loss: 0.1739 - acc: 0.9468
Epoch 2/5
60000/60000 [==============================] - 39s 652us/step - loss: 0.0457 - acc: 0.9859
Epoch 3/5
60000/60000 [==============================] - 36s 598us/step - loss: 0.0307 - acc: 0.9906
Epoch 4/5
60000/60000 [==============================] - 37s 614us/step - loss: 0.0239 - acc: 0.9930
Epoch 5/5
60000/60000 [==============================] - 35s 590us/step - loss: 0.0193 - acc: 0.9941
10000/10000 [==============================] - 2s 189us/step
test loss: 0.028174066650505848
test accuracy: 0.9913