(usage
The Sequential() method is a container for the
describing the network structure of a neural network.
Describe the network structure from the input layer to the output layer in the input parameter of Sequential()
model = ([network infrastructure]) #Describe the layers of the network
Examples of network structures
straighten layers
() #Straightening layers can change the size of the tensor,Straighten the input features into a one-dimensional array,is the layer without computational parameters
full connectivity layer
(Number of neurons, activation = "activation function“, kernel_regularizer = "regularization method)
Among them:
- Activation can be selected from relu, softmax, sigmoid, tanh, etc.
- kernel_regularizer optional .l1(), .l2()
convolutional layer
.Conv2D(filter = Number of convolution kernels, kernel_size = Convolution kernel size, strides = convolutional step, padding = ”valid“ or "same")
LSTM layer
()
sample code (computing)
# First step, import import tensorflow as tf #Import Module from sklearn import datasets #Importing datasets from sklearn import numpy as np #Import Scientific Computing Module import keras # Step two, train, test x_train = datasets.load_iris().data #Importing the inputs of an iris dataset y_train = datasets.load_iris().target #Importing labels for iris datasets (120) # Set random seeds so that the results are the same every time for easy cross-referencing (x_train) # Use the shuffle() method to disorganize the input x_train (120) # Set random seeds so that the results are the same every time for easy cross-referencing (y_train) # Use the shuffle() method to disorganize the input y_train .set_seed(120) #Let the number of seeds in tensorflow be set to 120 #Step 3, () model = ([ #Build a neural network using () (3, activation = "softmax", kernel_regularizer = .l2()) # Fully connected layer, three neurons, activation function is softmax, use l2 regularization ]) #Step 4, () ( # Use the () method to configure the training method optimizer = (lr = 0.1), #Using the SGD optimizer with a learning rate of 0.1 loss = (from_logits = False), #Configuration loss function metrics = ['sparse_categorical_accuracy'] #Labeling network evaluation metrics ) # Step 5, () ( # Use the () method to perform the training process. x_train, y_train, # Inform the training set of the inputs as well as the labels. batch_size = 32, # The size of each batch of batch is 32. epochs = 500, # of iterations epochs is 500 validation_split = 0.2, # Dividing 80% from the test set to the training set validation_freq = 20 # of test intervals 20 ) # Step 6, () () #Printing Neural Network Structures,Number of statistical parameters
summarize
The above is a personal experience, I hope it can give you a reference, and I hope you can support me more.