clear;
addpath(genpath([pwd '/DPC']));
addpath(genpath('D:/data/tree-group-lasso-screening'));
% This is an example for running the function tree_LeastR
%
% Problem:
%
% min 1/2 || A x - y||^2 + z * sum_j w_j ||x_{G_j}||
%
% G_j's are nodes with tree structure
%
% The tree structured group information is contained in
% opts.ind, which is a 3 x nodes matrix, where nodes denotes the number of
% nodes of the tree.
%
% opts.ind(1,:) contains the starting index
% opts.ind(2,:) contains the ending index
% opts.ind(3,:) contains the corresponding weight (w_j)
%
% Note:
% 1) If each element of x is a leaf node of the tree and the weight for
% this leaf node are the same, we provide an alternative "efficient" input
% for this kind of node, by creating a "super node" with
% opts.ind(1,1)=-1; opts.ind(2,1)=-1; and opts.ind(3,1)=the common weight.
%
% 2) If the features are well ordered in that, the features of the left
% tree is always less than those of the right tree, opts.ind(1,:) and
% opts.ind(2,:) contain the "real" starting and ending indices. That is to
% say, x( opts.ind(1,j):opts.ind(2,j) ) denotes x_{G_j}. In this case,
% the entries in opts.ind(1:2,:) are within 1 and n.
%
%
% If the features are not well ordered, please use the input opts.G for
% specifying the index so that
% x( opts.G ( opts.ind(1,j):opts.ind(2,j) ) ) denotes x_{G_j}.
% In this case, the entries of opts.G are within 1 and n, and the entries of
% opts.ind(1:2,:) are within 1 and length(opts.G).
%
%% Related papers
%
% [1] Jun Liu and Jieping Ye, Moreau-Yosida Regularization for
% Grouped Tree Structure Learning, NIPS 2010
%
%% load data
%profile on;
rng(1)
% ---------------------- generate random data ----------------------
N = 250;
p = 100000;
d = 3; % the depth of the tree excluding the root node
nns = [1,10,50,p]; % the size of each node at different levels
ratio = [1,0.2,0.5];
% data_name = ['syn1_' num2str(N) '_' num2str(p)];
% [ X, y, ind ] = gen_sync1( p, N, d, nns, ratio );
data_name = 'ADNI_WMV_full';
%[ X, y, beta, ind ] = gen_sync2( p, N, d, nns, ratio );
load('ADNI_TGL_full.mat');
y = WMV;
X = zscore(X);
y = zscore(y);
%% In this example, the tree is set as:
%
% root, 1:100, with weight 0
% its children nodes, 1:50, and 51:100
%
% For 1:50, its children are 1:20, 21:40, and 41:50
%
% For 51:100, its children are 51:70, and 71:100
%
% These nodes in addition have each individual features (they contain) as
% children nodes.
%
%%
%% One efficient way
% We make use of the fact that the indices of the left nodes of the tree
% are smaller than the right nodes.
%----------------------- Set optional items -----------------------
opts=[];
% Starting point
opts.init=1; % 2: starting from a zero point
% 1: warm start
% Termination
opts.tFlag=5; % run .maxIter iterations
opts.maxIter=1000; % maximum number of iterations
% regularization
opts.rFlag=1; % 1: use ratio
% 0: use the true value
% Normalization
opts.nFlag=0; % without normalization
% Group Property
opts.ind = ind;
% opts.ind=[[-1, -1, 1]',... % leave nodes (each node contains one feature)
% [1, 20, sqrt(20)]', [21, 40, sqrt(20)]',... % the layer above the leaf
% [41, 50, sqrt(10)]', [51, 70, sqrt(20)]', [71,90, sqrt(20)]', [91,100, sqrt(10)]',...
% [1, 50, sqrt(50)]', [51, 100, sqrt(50)]']; % the higher layer
%
% opts.depth = d;
% ----------------- set the parameter values ------------------
ub = 1;
lb = 0.05;
npar = 100;
scale = 'log';
Lambda = get_lambda(lb, ub, npar, scale);
%----------------------- Run the code tree_LeastR -----------------------
% --------------------- experiments ----------------
npar = length(Lambda);
rej_ratio = zeros(d,npar);
run_solver = 1;
run_time = [];
ntrial = 1;
fprintf('data: %s:\n\n',data_name);
for trial = 1:ntrial
fprintf('\ntrial %d:\n\n',trial);
%X = data_matrix;
%y = response;
% run the solver without screening
if run_solver == 1
if trial == 1
run_time.solver = 0;
run_time.solver_step = zeros(1,npar);
end
fprintf('computing the ground truth by solver...\n');
starts = tic;
[Sol_GT,funVal,tsol]=pathSolutionTGL(X, y, Lambda, opts); % ground truth solution
tsolver = toc(starts);
run_time.solver=run_time.solver + tsolver;
run_time.solver_step = run_time.solver_step + tsol;
fprintf('Running Time of solver without MLFre: %f\n',tsolver);
fprintf('Effective number of parameter values: %f\n',nnz(sum(Sol_GT.*Sol_GT)>=1e-20));
end
if opts.tFlag == 2
opts.funVal = funVal;
end
% Compute the solution with screening MLFre
if trial == 1
run_time.MLFre_solver = 0;
run_time.solver_MLFre_step = zeros(1,npar);
run_time.MLFre = 0;
end
fprintf('computing the solution with screening MLFre...\n');
starts = tic;
[Sol_MLFre, ind_zf_MLFre, ts_MLFre] = MLFre_TGL(X, y, Lambda, opts);
tMLFre_solver = toc(starts);
tMLFre = tMLFre_solver - sum(ts_MLFre);
run_time.MLFre_solver = run_time.MLFre_solver + tMLFre_solver;
run_time.solver_MLFre_step = run_time.solver_MLFre_step + ts_MLFre;
run_time.MLFre = run_time.MLFre + tMLFre;
fprintf('Running time of MLFre: %f\n',tMLFre);
fprintf('Running Time of solver with MLFre: %f\n',tMLFre_solver);
% --------------------- compute the rejection ratio --------------------- %
if run_solver == 1
Sol = Sol_GT;
fprintf('numerical error by MLFre: %f\n', ...
norm(Sol_GT-Sol_MLFre,'fro'));
else
Sol = Sol_MLFre;
end
ind_zf = abs(Sol)<1e-12;
rej_ratio = rej_ratio + reshape(sum(ind_zf_MLFre),d,npar)./repmat(sum(ind_zf),d,1);
end
if run_solver == 1
run_time.solver = run_time.solver / ntrial;
run_time.solver_step = run_time.solver_step / ntrial;
end
run_time.MLFre_solver = run_time.MLFre_solver / ntrial;
run_time.solver_MLFre_step = run_time.solver_MLFre_step / ntrial;
run_time.MLFre = run_time.MLFre / ntrial;
rej_ratio = rej_ratio / ntrial;
if run_solver == 1
speedup = run_time.solver/run_time.MLFre_solver;
fprintf('speedup: %f\n',speedup);
end
% ---------------------------- save the result ----------------------------
result_path = ['result/' data_name];
if ~exist(result_path,'dir')
mkdir(result_path);
end
result_path = [result_path '/' scale];
if ~exist(result_path,'dir')
mkdir(result_path);
end
result_path = [result_path '/'];
result_name = [data_name '_result_' scale];
if run_solver == 1
save([result_path result_name],'data_name','Lambda','rej_ratio','run_time','speedup');
else
save([result_path result_name],'data_name','Lambda','rej_ratio','run_time');
end
%% plot the rejection ratio
plot_rej_ratio(Lambda,rej_ratio,scale,data_name,result_path)
%profile viewer;
Datasets
Models
