demo_BNMC_blackbox_function.m

addpath('utilities\');

%fprintf('Loading Scene Data\n');
% [yyTrain, xxTrain]=read_sparse_ml('data\scene_train');
% [yyTest, xxTest]=read_sparse_ml('data\scene_test');
load('SceneData.mat');

eta_xx=0.1; % Dirichlet symmetric for SVI (feature)
eta_yy=0.01; % Dirichlet symmetric for SVI (label)
learn_rate=0.001; % learning rate for SVI
lambda=64;% for SGD
trun_thesh=0.00001; % truncation threshold
alpha=1; % stick-breaking parameter

parameters=[eta_xx,eta_yy,learn_rate,lambda,trun_thesh,alpha];

data.xxTrain=xxTrain;
data.yyTrain=yyTrain;
data.xxTest=xxTest;
data.yyTest=yyTest;


fprintf('eta_x=%.4f eta_y=%.4f learning_rate=%.4f lambda=%.1f truncation=%.6f stick-breaking=%.2f\n',eta_xx,eta_yy,learn_rate,lambda,trun_thesh,alpha);
Y=BayesOpt_BNMC(parameters,data);
fprintf('y=f(X)=%.3f\n',Y);