Keras Study (I.) _keras

I see that Keras is good, based on Python, the background is based on Theano or TensorFlow. Installation

Environment: ubuntu14.04
First, install the Python environment, Theano, and Keras

sudo apt-get install python-numpy python-scipy python-dev python-pip python-nose g++ the git
sudo pip libopenblas-dev All Theano
sudo pip install Keras

Data and Code Preparation

According to the blog, download Mnist.zip data and GitHub on the cnn.py and data.py, extract to the same directory:

Note: Keras default background is TensorFlow, changed to Theano, modify ~/.keras/keras.json file can be:

Image_dim_ordering and backend two parameters are modified (the default is TF and TensorFlow). Training CNN Model

Execute the Python cnn.py file to perform training:

Ran 10 epoch, accuracy rate nearly 99%.
Here's what CNN looks like with the code:

############### #开始建立CNN模型 ############### #生成一个model model = sequential () #第一个卷积层, 4 convolution cores, each convolution nucleus size 5*5.
1 represents the channel of the imported picture, and the grayscale image is 1 channels. #border_mode可以是valid或者full, specific look here Description: http://deeplearning.net/software/theano/library/tensor/nnet/conv.html# theano.tensor.nnet.conv.conv2d #激活函数用tanh #你还可以在model. Add (Activation (' tanh ') with dropout tips: model.add (Dropout (0.5 ) Model.add (convolution2d (4, 5, 5, border_mode= ' valid ', input_shape= (1,28,28))) Model.add (Activation (' Tanh ')) # The second convolution layer, 8 convolution cores, each volume kernel size 3*3. 4 represents the number of feature graphs entered, equal to the number of convolution cores on the previous layer #激活函数用tanh #采用maxpooling, poolsize (2,2) Model.add (convolution2d (8, 3, 3, border_mode= ' valid ') ) Model.add (Activation (' Tanh ')) Model.add (Maxpooling2d (pool_size= (2, 2)) #第三个卷积层, 16 convolution cores, each volume kernel size 3*3 #激活函数用tanh #
Using Maxpooling,poolsize (2,2) Model.add (convolution2d (3, 3, border_mode= ' valid ')) Model.add (Activation (' tanh '))
Model.add (Maxpooling2d (pool_size= (2, 2)) #全连接层, the first layer of the output of the two-dimensional feature map flatten as one-dimensional. #Dense就是隐藏层. 16 is the number of feature graphs that are printed on the previous layer. 4 is calculated according to each convolution layer: (28-5+1) gets 24, (24-3+1)/2 gets 11, (11-3+1)/2 gets 4 #全连接有128个神经元节点, beginningThe initialization mode is normal model.add (Flatten ()) Model.add (dense (128, init= ' normal ')) Model.add (Activation (' Tanh ')) #Softmax分类, Output is 10 category Model.add (dense (init= ' normal ')) Model.add (Activation (' Softmax ')) ############# #开始训练模型 ############## #
Using SGD + momentum #model. Compile parameters loss is the loss function (target function) SGD = SGD (lr=0.05, decay=1e-6, momentum=0.9, Nesterov=true) Model.compile (loss= ' categorical_crossentropy ', OPTIMIZER=SGD) #调用fit方法, is a training process.
The number of epoch for training is set to 10,batch_size 100. #数据经过随机打乱shuffle =true. verbose=1, the information output in the training process, 0, 1, 23 ways can be irrelevant.
Show_accuracy=true, every epoch is output accuracy during training.
#validation_split = 0.2, 20% of the data as a validation set.

 Model.fit (data, label, BATCH_SIZE=100, nb_epoch=10,shuffle=true,verbose=1,validation_split=0.2)

: http://blog.csdn.net/zhoubl668/article/details/45559955
Keras Chinese document: http://keras-cn.readthedocs.io/en/latest/