# Sparse Autoencoder (SAE)

DesignSAELayersize=[dim,5]; % design the structure of SAE, first and second element are visualsize(usually dimension of input data) and hiddensize (number of hidden note).
Option_SAE.ActivationFunction='sigmoid'; % 'sigmoid','linear','tanh','ReLU'
Option_SAE.LearningRate=0.05; % learning rate
Option_SAE.maxIter=1000; % maximum iteration
Option_SAE.lambda=0.0001; % weight decay parameter 
Option_Sparsity.beta=0.1;  % weight of sparsity penalty term 
Option_Sparsity.sparsityParam=0.01;  % desired average activation of the hidden units
LearningApproach='SGD';% 'SGD','Momentum','AdaGrad','RMSProp','Adam'; SGD is slow, but stable.

% initialize the SAE
Net=Initialization_Net_SAE(DesignSAELayersize,Option_SAE,Option_Sparsity,LearningApproach); 

Net.batchsize=10;
Net=SAE_Learning_batch(X,Net);