function [ FI ] = NicForest_FeatureImportance(f_out,xtrain,xtest)
%NICFOREST_FEATUREIMPORTANCE Calculate feature importance for a given
%ensemble model trained on xtrain. xtest must be an unused held-out data
%set.
%
% [ FI ] = NicForest_FeatureImportance(f_out,xtrain,xtest) calculates
% the feature importance as determined by the utility of its
% contributions in the ensemble of forests.
%
% Inputs:
% f_out - Bayesian ensemble output of NicForest_train
% xtrain - Training data used for forests
% xtest - Data not used in the training of forests
%
%
% Outputs:
% FI - 1xD feature importance values
%
%
% Example
% load PhysionetDataSetA
%
% xtrain=data(1:1500,:);
% ytrain=outcome(1:1500);
% xtest=data(1501:end,:);
% ytest=outcome(1501:end);
%
% load forest.mat
% [ fi ] = NicForest_FeatureImportance(forests,xtrain,xtest);
%
% See also NICFOREST_TRAIN APPLY_TREE TIEDRANKRELATIVE
% $LastChangedBy: alistair $
% $LastChangedDate: 2012-05-30 12:21:30 +0100 (Wed, 30 May 2012) $
% $Revision: 21 $
% Originally written on GLNXA64 by Alistair Johnson, 09-May-2012 14:58:24
% Contact: alistairewj@gmail.com
forests = f_out.forests;
%=== Rank test set according to train set
xtest_rk = tiedrankrelative([xtest;xtrain],...
[false(size(xtest,1),1);true(size(xtrain,1),1)]);
xtest_rk = xtest_rk(1:size(xtest,1),:);
%=== Calculate number of non-missing values in each feature and symmetrize
NaNNbr = sum( ~isnan(xtrain) , 1) + 1;
%=== Determine number of trees
NTrees = size(forests,3)-1;
% feature importance
FI = zeros(size(xtest));
for i=1:size(forests,1)
for j=1:NTrees
FI(:,forests(i,3,j)) = FI(:,forests(i,3,j)) + apply_tree( forests(i,:,j) , xtest_rk , NaNNbr );
end
if mod(i,10)==0
fprintf('%2.2f%% complete.\n',i/size(forests,1)*100);
end
end
FI = std(FI,[],1);
end