Online_DL.m

function[D,alpha,RS,t1,t2] =Online_DL(config_file)
%% Evaluate global configuration file
try
    eval(config_file);
catch
    disp('config file failed!')
end
%% load the feature_data
load([FEATURE_DIR,'data_tran.mat']);% use the standarlized features
%% descriptors for the sparse dictionary learning or Online sparse dictionary learning，only use several sequence for each class
temp_index = randperm(size(data_tran{1},2));%因为size=100,得到1到100的随机序列
all_descriptors = [];
for i = 1:size(data_tran,2)  %i从1到5
    for j = 1:size(temp_index,2)/18 % only use a subset (1/19 here) of training data to construct the codebook    j从1到5
        all_descriptors = [all_descriptors;data_tran{i}{temp_index(j)}];%下标为i和temp_index(j),第i个cell里的第temp_index(j)个
    end                                                                 %{temp_index(j)}是随机的，j是1到5；
end
param.K=100;  % learns a dictionary with 100 elements
param.lambda=0.001;
param.numThreads=4; % number of threads
param.batchsize=4000;
param.iter=1000;  % let us see what happens after 1000 iterations.

tic
D = mexTrainDL(all_descriptors',param);
t1=toc;
RS=[];
tic
for i=1:9;
    for j=1:231;
alpha{i}{j}=mexLasso(data_tran{i}{j}',D,param);
R=mean(0.5*sum((data_tran{i}{j}'-D*alpha{i}{j}).^2)+param.lambda*sum(abs(alpha{i}{j})));
RS=[RS;R];
end
end
t2=toc;
end