ML_GenerateModel_RF.m

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% This file constructs the model to be used for prediction by using the labelled images generated by LabelObjects_BATCH.m
% The SVM model can then be used by ML_PredictWithModel.m and ML_PredictWithModel_BATCH.m
% 2016-07-07 Romain Laine rfl30@cam.ac.uk
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% Initialization
clear all;
close all;
clc

% User defined parameters -------------------------------------------------

% Cross-validation
Cross_validation = 1;
Fold_for_CV = 10; % number of fold for cross-validation

Calculate_posterior_prob = 1;
N_sample_for_probability = 10;

% Model parameters
MaxNumSplits = 6000;
NumLearningCycles = 60;

Learners_names_selected = {'Area' 'AxisLengthRatio' 'Eccentricity' 'Solidity' 'Perim2Area_ratio' 'MeanIntensity' 'Std_pixel_values'...
    'Hu_IM1' 'Hu_IM4' 'Hu_IM5' 'Hu_IM6' 'Phi_4'...
    'AN1653' 'AN2859' 'AN4011' 'AN832' 'AN3486' 'AN2689'...
    'BoF22' 'BoF58' 'BoF95' 'BoF110' 'BoF177' 'BoF178'};

% Other options
Remove_unknown = 0;
Save_confusion_matrix = 1;
Save_model = 1;

Default_path = 'C:\Users\rfl30\DATA raw\SIM data\';


%% ------------------------------------------------------------------------
% Read in virus data with learners and classifiers

Model_parameters{1} = MaxNumSplits;
Model_parameters{2} = NumLearningCycles;

[Filename, Filepath] = uigetfile('*.xlsx','Choose an annotated descriptor file...',Default_path);

tic
disp('------------------------------');
disp('Reading data from spreadsheet...');
disp(fullfile(Filepath, Filename));
[Learners_values, annotation, ~] = xlsread(fullfile(Filepath, Filename),1);
toc

% Rename variables with the data and information
Learners_names = annotation(1,1:end-1);
annotation = annotation(2:end,end);
Learners_values = Learners_values(:,1:end);
n_learners = size(Learners_values,2);
disp(['Number of learners: ', num2str(n_learners)]);
disp(['Learners selected: ', Learners_names_selected]);
n_learners_selected = length(Learners_names_selected);

% Get the ordered list of selected learners
Learners_names_selected_ordered = Learners_names;
Learners_names_selected_ordered(~ismember(Learners_names, Learners_names_selected)) = [];

if n_learners_selected ~= length(Learners_names_selected_ordered)
    disp('Learners list in .xls file not matching those selected.');
    return;
end

%%

if Remove_unknown == 1
    % Remove all unknown data points
    inds = ~strcmp(annotation,'Unknown');
    Learners_values = Learners_values(inds,:);
    annotation = annotation(inds);
end

% Get the list of classes
Class_list = unique(annotation);
n_classes = length(Class_list);
disp(['Number of classes: ',num2str(n_classes)]);
disp('Class list:');
disp(Class_list);

% Class_list{1} = 'Filamentous';
% Class_list{2} = 'Small filamentous';
% Class_list{3} = 'Large spherical';
% Class_list{4} = 'Small spherical';
% Class_list{5} = 'Rod';
% Class_list{6} = 'Unknown';


n_examples = size(annotation,1);
disp(['Number of examples: ', num2str(n_examples)]);

% Select the learners
Learners_values_selected = Learners_values(:,ismember(Learners_names, Learners_names_selected_ordered));

trainedClassifier = GenerateModel_RandomForest(Learners_values_selected, annotation, ...
    Model_parameters, [], Class_list, Learners_names_selected_ordered);

%% ------------------------------------------------------------------------
% Choose a name for the model
if Save_model == 1
    RF_model_name_default = 'RF_Model';
    answer = inputdlg('Choose the file name for the model','Model name:',1,{RF_model_name_default});
    disp('------------------------------');
    save(fullfile(Filepath,[answer{1},'.mat']), 'trainedClassifier');
    disp('RF model saved as:');
    disp(fullfile(Filepath,[answer{1},'.mat']));
end

%%

if Cross_validation == 1
    %% Perform cross-validation
    tic
    disp('------------------------------');
    disp('Performing cross-validation...');
    partitionedModel = crossval(trainedClassifier.ClassificationEnsemble, 'KFold', Fold_for_CV);
    % Compute validation accuracy
    validationAccuracy = 1 - kfoldLoss(partitionedModel, 'LossFun', 'ClassifError');
    disp(['Accuracy: ',num2str(100*validationAccuracy,'%.1f'),' %']);
    
    % Compute validation predictions and scores
    validationPredictions = kfoldPredict(partitionedModel);
    [C,order] = confusionmat(categorical(annotation), validationPredictions);
    toc
    
    % Convert the integer label vector to a class-identifier matrix.
    [~,grp] = ismember(validationPredictions, Class_list);
    oofLabelMat = zeros(n_classes, n_examples);
    idxLinear = sub2ind([n_classes n_examples],grp,(1:n_examples)');
    oofLabelMat(idxLinear) = 1; % Flags the row corresponding to the class
    YMat = zeros(n_classes, n_examples);
    [~,grp] = ismember(annotation, Class_list);
    idxLinearY = sub2ind([n_classes n_examples],grp,(1:n_examples)');
    YMat(idxLinearY) = 1;
    
    %%
%     Class_list2 = cell(6,1);
%     Class_list2{1} = 'LF';
%     Class_list2{2} = 'SF';
%     Class_list2{3} = 'LS';
%     Class_list2{4} = 'SS';
%     Class_list2{5} = 'RD';
%     Class_list2{6} = 'UK';
    
    h_ConfMat = figure('Color','white','name','Confusion matrix');
    hPC = plotconfusion(YMat,oofLabelMat);
    h = gca;
    h.XTickLabel = [Class_list; {''}];
    h.YTickLabel = [Class_list; {''}];
    %%
    
    if Save_confusion_matrix == 1
        % Save the confusion matrix
        conf_matrix.matrix = C;
        conf_matrix.order = order;
        disp('------------------------------');
        disp('Saving confusion matrix...');
        %         save(fullfile(Filepath,[answer{1},'_conf_matrix.mat']), 'conf_matrix');
        saveas(h_ConfMat, fullfile(Filepath,[answer{1},'_conf_matrix image.png']), 'png');
    end
    
    %%
    if Calculate_posterior_prob == 1
        [label, Posterior] = resubPredict(trainedClassifier.ClassificationEnsemble);
        idx = randsample(size(Learners_values,1),N_sample_for_probability,0);
        %         SVMModel.ClassNames
        Probability_table = table(annotation(idx),label(idx),Posterior(idx,:),...
            'VariableNames',{'TrueLabel','PredLabel','Posterior'});
        disp(Probability_table);
    end
    
end

disp('-----------------------------');
disp('All done.');