function [Pz_d_test,labels] = do_plsa_test(config_file)

%% Test plsa model

%% Runs the pLSA procedure, this time holding P(w|z) constant and
%% estimating P(d|z) using a "folding in" procedure. Calls code written
%% by Josef Sivic (josef@robots.ox.ac.uk), based on the paper:
%% 
%%  J. Sivic, B. C. Russell, A. Efros, A. Zisserman and W. T. Freeman,
%%  Discovering objects and their location in images, ICCV 2005.

%% The action of this routine depends on the directory in which it is
%% run: 
%% (a) If run from RUN_DIR, then it will evaluate the latest model in the
%% models subdirectory. i.e. if you have just run
%% do_plsa('config_file_2'), which saved to model_0011.mat and
%% config_file_0011.m in the models subdirectory in RUN_DIR, then doing
%% do_plsa_evaluation('config_file_2') will load up model_0011.mat and
%% evaluate it. 
%% (b) If run within in models subdirectory, then it
%% will evaluate the model corresponding to the configuration file passed
%% to it. i.e. do_plsa_evaluation('config_file_0002') will load
%% model_0002.mat and evaluate/plot figures for it. 
%%  
%% Mode (a) exists to allow a complete experiment to be run from start to
%% finish without having to manually go into the models subdirectory and
%% find the appropriate one to evaluate.
  
%% If this routine is called on a newly learnt model, it will run the pLSA code
%% in folding in mode and then plot lots of figures. If run a second time
%% on the same model, it will only plot the figures, since there is no need
%% to recompute the P(d|z) on the testing images. If you want to force it
%% to re-run on the images, then remove the Pd_z_test variable from the
%% model file. 
 
  
%% R.Fergus (fergus@csail.mit.edu) 03/10/05.  
 
  
%% Evaluate global configuration file
eval(config_file);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%% Model section
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%% get filename of model to learn
%%% if in models subdirectory then just get index off config_file string
if (strcmp(pwd,[RUN_DIR,'/',Global.Model_Dir_Name]) | strcmp(pwd,[RUN_DIR,'\',Global.Model_Dir_Name]))
    ind = str2num(config_file(end-Global.Num_Zeros+1:end));
else
    %%% otherwise just take newest model in subdir.
    ind = length(dir([RUN_DIR,'/',Global.Model_Dir_Name,'/',Global.Model_File_Name,'*.mat']));    
end
%%% construct model file name
model_fname = [RUN_DIR,'/',Global.Model_Dir_Name,'/',Global.Model_File_Name,prefZeros(ind,Global.Num_Zeros),'.mat'];

%%% load up model
load(model_fname);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Test section - run model on testing images only if Pd_z_test does not exist
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

if ~exist('Pd_z_test') %%% only do this section the first time we look at the model
                       %%% saves time if we just want to look at the pretty
                       %%% figures
    
    %% get all file names of testing image interest point files.
    ip_file_names =  genFileNames({Global.Interest_Dir_Name},Categories.All_Test_Frames,RUN_DIR,Global.Interest_File_Name,'.mat',Global.Num_Zeros);

    %% Create matrix to hold word histograms from all images
    X = zeros(VQ.Codebook_Size,length(Categories.All_Test_Frames));

    %% load up all interest_point files which should have the histogram
    %% variable already computed (performed by do_vq routine).
    for a=1:length(ip_file_names)
        %% load file
        load(ip_file_names{a});
        %% store histogram
        X(:,a) = histogram';    
    end 

    %%% Call actual EM routine again in test mode
    [Pw_z,Pd_z_test,Pz_test,Li_test] = pLSA_EM(X,Pw_z,Learn.Num_Topics,Learn);
    Pdz_test = Pd_z_test.*repmat(Pz_test,100,1);
    Pd_test = sum(Pdz_test,2);
    Pz_d_test = Pdz_test./repmat(Pd_test,1,2);
    
    %%% get labels for test frames
    labels = [];
    for a=1:Categories.Number
        labels = [labels , Categories.Labels(a)*ones(1,length(Categories.Test_Frames{a}))];
    end
   
    %%% compute classification performance for each topic
    for t=1:Learn.Num_Topics
        %%% get scores for each image
        values = Pd_z_test(:,t)';

    end 


end