mil_run.m

包含了很多分类算法

%     msg = sprintf(' Coding: ECOC63_31 ');
%     preprocess.Message = [preprocess.Message msg]; 
% elseif (preprocess.MultiClass.CodeType == 4)
%     msg = sprintf(' Coding: Random ');
%     preprocess.Message = [preprocess.Message msg]; 
% end;    

% if (preprocess.MultiClass.LossFuncType == 0) 
%     msg = sprintf(' Loss: L1 ');
%     preprocess.Message = [preprocess.Message msg];
% elseif (preprocess.MultiClass.LossFuncType == 1)
%     msg = sprintf(' Loss: Exp ');
%     preprocess.Message = [preprocess.Message msg]; 
% elseif (preprocess.MultiClass.LossFuncType == 2)
%     msg = sprintf(' Loss: (1-Y)+ ');
%     preprocess.Message = [preprocess.Message msg];
% end;    
% 
% if (preprocess.MultiClass.UncertaintyFuncType == 0) 
%     msg = sprintf(' Uncertainty: L1 ');
%     preprocess.Message = [preprocess.Message msg];
% elseif (preprocess.MultiClass.UncertaintyFuncType == 1)
%     msg = sprintf(' Uncertainty: Exp ');
%     preprocess.Message = [preprocess.Message msg]; 
% elseif (preprocess.MultiClass.UncertaintyFuncType == 2)
%     msg = sprintf(' Uncertainty: (1-Y)+ ');
%     preprocess.Message = [preprocess.Message msg];
% elseif (preprocess.MultiClass.UncertaintyFuncType == 3)
%     msg = sprintf(' Uncertainty: Min Margin ');
%     preprocess.Message = [preprocess.Message msg];
% elseif (preprocess.MultiClass.UncertaintyFuncType == 4)
%     msg = sprintf(' Uncertainty: -ln(1+x) ');
%     preprocess.Message = [preprocess.Message msg];
% elseif (preprocess.MultiClass.UncertaintyFuncType == 5)
%     msg = sprintf(' Uncertainty: Random ');
%     preprocess.Message = [preprocess.Message msg];
% end;    
% 
% if (preprocess.MultiClass.ProbEstimation == 0) 
%     msg = sprintf(' Best Worse ');
%     preprocess.Message = [preprocess.Message msg];
% elseif (preprocess.MultiClass.ProbEstimation == 1)
%     msg = sprintf(' Uniform Guess ');
%     preprocess.Message = [preprocess.Message msg]; 
% elseif (preprocess.MultiClass.ProbEstimation == 2)
%     msg = sprintf(' Binary Class ');
%     preprocess.Message = [preprocess.Message msg];     
% end;

% if (preprocess.DataSampling == 1)
%     msg = sprintf(' Sampling Rate: %d', preprocess.DataSamplingRate);
%     preprocess.Message = [preprocess.Message msg];    
% end;

% num_folder = preprocess.NumFolder;
% num_cross_folder = preprocess.NumCrossFolder;; 

% load in the data
if ((~isfield(preprocess, 'input_file')) | (isempty(preprocess.input_file))), 
    error('The input file is not provided!');
end;

input_file = preprocess.input_file;
output_file = preprocess.output_file;
pred_file = preprocess.pred_file;

fid = fopen(output_file, preprocess.OutputFlag);
if (fid < 0),
    if ((~isfield(preprocess, 'test_file')) | (isempty(preprocess.test_file))), 
        output_file = sprintf('%s.result', input_file);
    else
        output_file = sprintf('%s.result', preprocess.test_file);
    end;
    preprocess.output_file = output_file;
    fid = fopen(output_file, preprocess.OutputFlag);
end;
if (preprocess.OutputFormat == 0),
    fprintf(fid, '\nProcessing Filename: %s\n', preprocess.input_file); 
    fprintf(fid, 'Classifier:%s\nMessage:%s\n',classifier, preprocess.Message);
end;
fclose(fid);    

fid = fopen(pred_file, preprocess.OutputFlag);
if (fid < 0),
    if ((~isfield(preprocess, 'test_file')) | (isempty(preprocess.test_file))), 
        pred_file = sprintf('%s.pred', input_file);
    else
        pred_file = sprintf('%s.pred', preprocess.test_file);
    end;
    preprocess.pred_file = pred_file;
    fid = fopen(pred_file, preprocess.OutputFlag);
end;
if (preprocess.PredFormat == 0),
    fprintf(fid, '\nProcessing Filename: %s\n', preprocess.input_file); 
    fprintf(fid, 'Classifier:%s\nMessage:%s\n',classifier, preprocess.Message);
end;
fclose(fid);    

% in MIL, we use a function to read the input data and convert into bag
% formats; sparse format is not currently supported 

% if (preprocess.InputFormat == 1),
%     strcmd = sprintf('!perl %s/ConvertFileInput.pl %s %s.stdout', root, input_file, input_file);
%     fprintf('!perl %s/ConvertFileInput.pl %s %s.stdout\n', root, input_file, input_file); 
%     eval(strcmd);
%     D = dlmread(sprintf('%s.stdout', input_file));
% else
%     D = dlmread(input_file);
% end;
% 
% if (preprocess.Sparse == 1),
%     D = spconvert(D);
% end;

% Automatically judge whether the shot information is available
% if (preprocess.ShotAvailable < 0),
%    if (length(unique(D(:, size(D, 2)))) > 10), 
%        preprocess.ShotAvailable = 1;
%    else
%        preprocess.ShotAvailable = 0;
%    end;
%    fprintf('Automatically detect preprocess.ShotAvailable to be %d \n', preprocess.ShotAvailable );
% end;

fprintf('Finished loading %s.............\n', input_file);
fprintf('Output Results to %s.............\n', output_file);
fprintf('Output Predictions to %s.............\n', pred_file);

switch (preprocess.Evaluation)
    case 0
        % Train Test
        EvaluationHandle = @MIL_Train_Test_Validate;
    case 1
        % Cross Validate
        EvaluationHandle = @MIL_Cross_Validate;
    case 2
        % Test File Validate
        if ((~isfield(preprocess, 'test_file')) | (isempty(preprocess.test_file))), 
            error('The test file is not provided!');
        end;
       %D_test = dlmread(preprocess.test_file, ',');
       %fprintf('Finished loading the test file %s.............\n', preprocess.test_file);
       %preprocess.TrainTestSplitBoundary = size(D, 1);
        preprocess.Shuffled = 0;
        %D = [D; D_test];
        EvaluationHandle = @MIL_Train_Test_Validate;   
    case 3
        % Training Only
        EvaluationHandle = @MIL_Train_Validate;
    case 4
        % Testing Only
        EvaluationHandle = @MIL_Test_Validate;
    case 5
        EvaluationHandle = @MIL_Leave_One_Out;           
        preprocess.Shuffled = 0;
end;

fhandle = @MIL_Train_Test_Simple;   %currently only address the binary MIL classification probleem 
% switch (preprocess.MultiClassType)
%     case 0
%         fhandle = @Train_Test_Simple;
%     case 1
%         fhandle = @Train_Test_Multiple_Class;
%     case 2
%         fhandle = @Train_Test_Multiple_Label;
%     case 3
%         fhandle = @Train_Test_Multiple_Class_AL;
% end;
       
fprintf('Classifier:%s\nMessage:%s\n',classifier, preprocess.Message);
if ((preprocess.Evaluation >= 3) & (preprocess.Evaluation <= 4))
    run = feval(EvaluationHandle, input_file, classifier);
else
    run = feval(EvaluationHandle, input_file, fhandle, classifier);
end;

OutputResult = [];
if (isfield(run, 'BagAccu')), OutputResult = [OutputResult sprintf('Bag label accuracy = %f, ', run.BagAccu)]; end;
if (isfield(run, 'InstAccu')), OutputResult = [OutputResult sprintf('Instance label accuracy = %f, ', run.InstAccu)]; end;

%if (isfield(run, 'Err')), OutputResult = [OutputResult sprintf('Error = %f, ', run.Err)]; end;
%if (isfield(run, 'Prec')), OutputResult = [OutputResult sprintf('Precision = %f, ', run.Prec)]; end;
%if (isfield(run, 'Rec')), OutputResult = [OutputResult sprintf('Recall = %f, ', run.Rec)]; end;
%if (isfield(run, 'F1')), OutputResult = [OutputResult sprintf('F1 = %f, ', run.F1)]; end;
%if (isfield(run, 'Micro_Prec')), OutputResult = [OutputResult sprintf('Micro_Precision = %f, ', run.Micro_Prec)]; end;
%if (isfield(run, 'Micro_Rec')), OutputResult = [OutputResult sprintf('Micro_Recall = %f, ', run.Micro_Rec)]; end;
%if (isfield(run, 'Micro_F1')), OutputResult = [OutputResult sprintf('Micro_F1 = %f, ', run.Micro_F1)]; end;
%if (isfield(run, 'Macro_Prec')), OutputResult = [OutputResult sprintf('Macro_Precision = %f, ', run.Macro_Prec)]; end;
%if (isfield(run, 'Macro_Rec')), OutputResult = [OutputResult sprintf('Macro_Recall = %f, ', run.Macro_Rec)]; end;
%if (isfield(run, 'Macro_F1')), OutputResult = [OutputResult sprintf('Macro_F1 = %f, ', run.Macro_F1)]; end;
%if (isfield(run, 'AvgPrec')), OutputResult = [OutputResult sprintf('MAP = %f, ', run.AvgPrec)]; end;
%if (isfield(run, 'BaseAvgPrec')), OutputResult = [OutputResult sprintf('MBAP = %f, ', run.BaseAvgPrec)]; end;

fprintf('%s\n', OutputResult);
if (~isempty(output_file)) 
    if ((preprocess.OutputFormat == 0) | (preprocess.OutputFormat == 1)),
        fid = fopen(output_file, 'a');
        fprintf(fid, '%s\n', OutputResult);
        fclose(fid);    
    end;
end;
if (~isempty(pred_file))
    fid = fopen(pred_file, 'w');
    fprintf(fid, 'Testing Bag Label Evaluation:\n');
    fprintf(fid, 'Index\tProbability\tPredict\tTruth\n');
    fprintf(fid, '%d\t%g\t%d\t%d\n', run.bag_pred');
    
%     if (preprocess.OutputFormat == 0),
%         if (size(run.Y_pred, 2) == 4),
%             fprintf(fid, 'Index\tProb\t\tPred\tTruth\n');
%             fprintf(fid, '%d\t%f\t%d\t%d\n', run.Y_pred');
%         else
%             fprintf(fid, 'Index\tProb\t\tPred\tTruth\tShotID\n');
%             fprintf(fid, '%d\t%f\t%d\t%d\t%d\n', run.Y_pred');
%         end;
%     elseif (preprocess.OutputFormat == 1), % Only work when number of classes is 2
%         fprintf(fid, 'Label: '); fprintf(fid, '%d ', preprocess.OrgClassSet); fprintf(fid, '\n');
%         ProbClass1 = run.Y_pred(:, 2) .* (run.Y_pred(:, 3) == 1)  + (1 - run.Y_pred(:, 2)) .* (run.Y_pred(:, 3) ~= 1);
%         fprintf(fid, '%d\t%f\t%f\n', [preprocess.OrgClassSet(run.Y_pred(:, 3)) ProbClass1 (1 - ProbClass1)]');
%     end;
    fclose(fid);
end;

function Report_Error()
fprintf(' Example: Suppose you are in root directory\n');
fprintf(' Example: test_classify.exe \"classify -t demo/DataExample1.train.txt -sh 1 -- train_only -m demo/DataExample1.libSVM.model -- LibSVM -Kernel 0 -CostFactor 3\" \n');
fprintf(' Example: test_classify.exe \"classify -t demo/DataExample1.test.txt -sh 1 -- test_only -m demo/DataExample1.libSVM.model -- LibSVM -Kernel 0 -CostFactor 3\" \n');
fprintf(' Please refer to http://finalfantasyxi.inf.cs.cmu.edu/tmp/MATLABArsenal.htm for more examples \n');
error('  The command must begin with Classify     ');