Building neural network based on tensorflow

be based on tensorflow Building neural networks

One 、tf.keras Steps of building neural network

Six steps

1. import
2. train,test
3. model = tf.keras.models.Sequential
4. model.compile
5. model.fit
6. model.summary

models.Sequential()

model = tf.keras.models.Sequential ([ Network structure ]) # Describe each layer of network
Network structure example ：

1. Straightening layer ：

tf.keras.layers.Flatten( )

2. Fully connected layer ：

tf.keras.layers.Dense( Number of neurons , activation= “ Activation function ”,kernel_regularizer= What kind of regularization )

activation（ The string gives ） Optional : relu、 softmax、 sigmoid 、 tanh
kernel_regularizer Optional : tf.keras.regularizers.l1()、tf.keras.regularizers.l2()

3. Convolution layer ：

tf.keras.layers.Conv2D(filters = Number of convolution kernels , kernel_size = Convolution kernel size ,strides = Convolution step , padding = " valid" or “same”)

4. LSTM layer ： tf.keras.layers.LSTM()

model.compile()

model.compile(optimizer = Optimizer ,
loss = Loss function
metrics = [“ Accuracy rate ”] )

Optimizer Optional :
‘sgd’ or tf.keras.optimizers.SGD (lr= Learning rate ,momentum= Momentum parameter )
‘adagrad’ or tf.keras.optimizers.Adagrad (lr= Learning rate )
‘adadelta’ or tf.keras.optimizers.Adadelta (lr= Learning rate )
‘adam’ or tf.keras.optimizers.Adam (lr= Learning rate , beta_1=0.9, beta_2=0.999)

loss Optional :
‘mse’ or tf.keras.losses.MeanSquaredError()
‘sparse_categorical_crossentropy’ or tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False)

Metrics Optional :
‘accuracy’ ：y_ and y It's all numbers , Such as y_=[1] y=[1]
‘categorical_accuracy’ ：y_ and y It's all hot code ( A probability distribution ), Such as y_=[0,1,0] y=[0.256,0.695,0.048]
‘sparse_categorical_accuracy’ ：y_ Numerical value ,y It's the only hot code ( A probability distribution ), Such as y_=[1] y=[0.256,0.695,0.048]

model.fit()

model.fit ( Input characteristics of training set , Training set label ,
batch_size= , epochs= ,
validation_data=( Input characteristics of validation set , Label of validation set ),
validation_split= How much proportion is divided from the training set to the verification set ,
validation_freq = How many times epoch Verify once )

model.summary()

Two 、 Customize Model

class MyModel(Model):
    def __init__(self):
        super(MyModel, self).__init__()
         Define the network structure block 
    def call(self, x):
         Call the network structure block , Achieve forward propagation 
        return y
model = MyModel()

example ：

class IrisModel(Model):
    def __init__(self):
        super(IrisModel, self).__init__()
        self.d1 = Dense(3, activation='softmax', kernel_regularizer=tf.keras.regularizers.l2())

    def call(self, x):
        y = self.d1(x)
        return y

model = IrisModel()

3、 ... and 、tf.keras Realize handwritten numeral classification

import tensorflow as tf

mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              metrics=['sparse_categorical_accuracy'])

model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1)
model.summary()

The operation results are as follows ：

60000/60000 [==============================] - 5s 89us/sample - loss: 0.0455 - sparse_categorical_accuracy: 0.9861 - val_loss: 0.0806 - val_sparse_categorical_accuracy: 0.9752
Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param # 
=================================================================
flatten (Flatten)            multiple                  0         
_________________________________________________________________
dense (Dense)                multiple                  100480    
_________________________________________________________________
dense_1 (Dense)              multiple                  1290      
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0