test_classify.m

一个matlab的工具包,里面包括一些分类器 例如 KNN KMEAN SVM NETLAB 等等有很多.

function run = test_classify(classifier)

warning('off','MATLAB:colon:operandsNotRealScalar');

% clear global preprocess;
global preprocess; 
global temp_train_file temp_test_file temp_output_file temp_model_file weka_dir mySVM_dir libSVM_dir SVMLight_dir; 
preprocess = [];

if (~isfield(preprocess, 'Message')), preprocess.Message = ''; end;
if (~isfield(preprocess, 'NumCrossFolder')), preprocess.NumCrossFolder = 3; end;
if (~isfield(preprocess, 'TrainTestSplitBoundary')), preprocess.TrainTestSplitBoundary = 100; end;
if (~isfield(preprocess, 'Normalization')), preprocess.Normalization = 1; end;
if (~isfield(preprocess, 'SizeFactor')), preprocess.SizeFactor = 0.5; end;
if (~isfield(preprocess, 'ShotAvailable')), preprocess.ShotAvailable = 0; end;
if (~isfield(preprocess, 'DataSampling')), preprocess.DataSampling = 0; end;
if (~isfield(preprocess, 'Sparse')), preprocess.Sparse = 0; end;
if (~isfield(preprocess, 'Shuffled')), preprocess.Shuffled = 0; end;
if (~isfield(preprocess, 'OutputFlag')), preprocess.OutputFlag = 'a'; end;
if (~isfield(preprocess, 'SVD')), preprocess.SVD = 0; end;
if (~isfield(preprocess, 'FLD')), preprocess.FLD = 0; end;
if (~isfield(preprocess, 'CHI')), preprocess.ChiSquare = 0; end;
if (~isfield(preprocess, 'ValidateByShot')), preprocess.ValidateByShot = 0; end;
if (~isfield(preprocess, 'Ensemble')), preprocess.Ensemble = 0; end;
if (~isfield(preprocess, 'ComputeMAP')), preprocess.ComputeMAP = 0; end;
if (~isfield(preprocess, 'Evaluation')), preprocess.Evaluation = 0; preprocess.TrainTestSplitBoundary = -2; end;
if (~isfield(preprocess, 'MultiClassType')), preprocess.MultiClassType = 0; end;
if (~isfield(preprocess, 'MultiClass') | (preprocess.MultiClassType == 0)), 
    preprocess.MultiClass.LabelType = 1; preprocess.MultiClass.CodeType = -1; preprocess.MultiClass.LossFuncType = -1;
    preprocess.MultiClass.UncertaintyFuncType = -1; preprocess.MultiClass.ProbEstimation = -1;
end;
if (~isfield(preprocess, 'ConstraintAvailable')), preprocess.ConstraintAvailable = 0; end;
if (~isfield(preprocess, 'ConstraintFileName')), preprocess.ConstraintFileName = ''; end;
if (~isfield(preprocess, 'input_file')), preprocess.input_file = ''; end;
if (~isfield(preprocess, 'output_file')), preprocess.output_file = ''; end;
if (~isfield(preprocess, 'pred_file')), preprocess.pred_file = ''; end;
if (~isfield(preprocess, 'model_file')), preprocess.model_file = ''; end;
if (~isfield(preprocess, 'WorkingDir')), preprocess.WorkingDir = ''; end;

if (nargin < 1), Report_Error; end;
[header, para, rem] = ParseCmd(classifier, '--');
if (strcmpi(header, 'classify')), 
    p = str2num(char(ParseParameter(para, {'-v'; '-sf'; '-n'; '-sh'; '-vs'; '-ds'; '-dsr'; '-svd'; '-fld';'-map'; '-if'; '-chi'; '-of'; '-pf'; '-sp'}, ...
                                          {'1'; '0'; '1'; '-1'; '1'; '0'; '0'; '0'; '0'; '0'; '0'; '0'; '0'; '0'; '0'})));
    preprocess.Vebosity = p(1);
    preprocess.Shuffled = p(2);
    preprocess.Normalization = p(3);
    preprocess.ShotAvailable = p(4);
    preprocess.ValidateByShot = p(5);
    preprocess.DataSampling = p(6);
    preprocess.DataSamplingRate = p(7);
    preprocess.SVD = p(8);
    preprocess.FLD = p(9);
    preprocess.ComputeMAP = p(10);
    preprocess.InputFormat = p(11);
    preprocess.ChiSquare = p(12);
    preprocess.OutputFormat = p(13);
    preprocess.PredFormat = p(14);
    preprocess.Sparse = p(15);
    p = ParseParameter(para, {'-t'; '-o'; '-p'; '-oflag'; '-dir'; '-drf' }, {''; ''; ''; 'a'; ''; ''});
    preprocess.input_file = char(p{1, :});
    preprocess.output_file = char(p{2, :});    
    preprocess.pred_file = char(p{3, :});    
    preprocess.OutputFlag = char(p{4, :});    
    preprocess.WorkingDir = char(p{5, :});    
    preprocess.DimReductionFile = char(p{6, :});    
    classifier = rem;   
else 
    Report_Error;
end;

% Setup the environmental varaible for directory information
if (isempty(preprocess.WorkingDir)), 
    % preprocess.WorkingDir = cd;
    filename = 'AggregatePredByShot.m'; 
    if (~exist(filename)),
        error('Cannot find the files in MATLABArsenal!');
    end;
    cur_dir = which(filename); 
    sep_pos = findstr(cur_dir, filesep); 
    preprocess.WorkingDir = cur_dir(1:sep_pos(length(sep_pos))-1);
end;
root = preprocess.WorkingDir;

temp_dir = sprintf('%s/temp', root);
if (~exist(temp_dir)),
    % eval(sprintf('!md \"%s\"', temp_dir));
    s = mkdir(root, 'temp');
    if (s ~= 1), error('Cannot create temp directory!'); end;
end;
temp_train_file = sprintf('%s/temp.train.txt', temp_dir);
temp_test_file = sprintf('%s/temp.test.txt', temp_dir);
temp_output_file = sprintf('%s/temp.output.txt', temp_dir);
temp_model_file = sprintf('%s/temp.model.txt', temp_dir);
weka_dir = sprintf('%s/weka-3-4/weka.jar', root);
mySVM_dir = sprintf('%s/svm', root);
libSVM_dir = sprintf('%s/svm', root);
SVMLight_dir = sprintf('%s/svm', root);

[header, para, rem] = ParseCmd(classifier, '--');
if (strcmpi(header, 'train_test_validate')), 
    preprocess.Evaluation = 0;
    p = str2num(char(ParseParameter(para, {'-t'}, {'-2'})));
    preprocess.TrainTestSplitBoundary = p(1);
    classifier = rem;
elseif (strcmpi(header, 'cross_validate')), 
    preprocess.Evaluation = 1;
    p = str2num(char(ParseParameter(para, {'-t'}, {'3'})));
    preprocess.NumCrossFolder = p(1);
    classifier = rem;
elseif (strcmpi(header, 'test_file_validate')), 
    preprocess.Evaluation = 2;
    p = char(ParseParameter(para, {'-t'}, {''}));
    preprocess.test_file = p(1, :);
    classifier = rem;
elseif (strcmpi(header, 'train_only')), 
    preprocess.Evaluation = 3;
    p = char(ParseParameter(para, {'-m'}, {''}));
    preprocess.model_file = p(1, :);
    temp_model_file = preprocess.model_file;
    classifier = rem;
elseif (strcmpi(header, 'test_only')), 
    preprocess.Evaluation = 4;
    p = char(ParseParameter(para, {'-m'}, {''}));
    preprocess.model_file = p(1, :);
    temp_model_file = preprocess.model_file;
    classifier = rem;
end;   

[header, para, rem] = ParseCmd(classifier, '--');
if (strcmpi(header, 'train_test_simple')), 
    preprocess.MultiClassType = 0;
    p = str2num(char(ParseParameter(para, {'-LabelType'}, {'1'})));
    preprocess.MultiClass.LabelType = p(1);
    classifier = rem;
elseif (strcmpi(header, 'train_test_multiple_class')), 
    preprocess.MultiClassType = 1;
    p = str2num(char(ParseParameter(para, {'-LabelType'; '-CodeType'; '-LossFuncType'}, {'1'; '0'; '2'})));
    preprocess.MultiClass.LabelType = p(1);
    preprocess.MultiClass.CodeType = p(2);
    preprocess.MultiClass.LossFuncType = p(3);
    preprocess.MultiClass.UncertaintyFuncType = 2; 
    preprocess.MultiClass.ProbEstimation = 0;
    classifier = rem;
elseif (strcmpi(header, 'train_test_multiple_label')), 
    preprocess.MultiClassType = 2;
    p = str2num(char(ParseParameter(para, {'-LabelType'}, {'1'})));
    preprocess.MultiClass.LabelType = p(1);
    classifier = rem;
elseif (strcmpi(header, 'train_test_multiple_class_AL')), 
    preprocess.MultiClassType = 3;
    p = str2num(char(ParseParameter(para, {'-LabelType'; '-CodeType'; '-LossFuncType'; '-ALIter'; '-ALIncrSize'}, {'1'; '0'; '2'; '4'; '10'})));
    preprocess.MultiClass.LabelType = p(1);
    preprocess.MultiClass.CodeType = p(2);
    preprocess.MultiClass.LossFuncType = p(3);
    preprocess.ActiveLearning.Iteration = p(4);
    preprocess.ActiveLearning.IncrementSize = p(5);
    preprocess.MultiClass.UncertaintyFuncType = 2; 
    preprocess.MultiClass.ProbEstimation = 0;
    classifier = rem;
end;   

% Initialize the message string

preprocess.Message = '';
if (preprocess.Evaluation == 0)
    msg = sprintf(' Train-Test Split, Boundary: %d, ', preprocess.TrainTestSplitBoundary);
    preprocess.Message = [preprocess.Message msg]; 
elseif (preprocess.Evaluation == 1)
    msg = sprintf(' Cross Validation, Folder: %d, ', preprocess.NumCrossFolder);
    preprocess.Message = [preprocess.Message msg];     
elseif (preprocess.Evaluation == 2)
    msg = sprintf(' Testing on File %s, ', preprocess.test_file);
    preprocess.Message = [preprocess.Message msg];     
elseif (preprocess.Evaluation == 3)
    msg = sprintf(' Training on File %s, ', preprocess.input_file);
    preprocess.Message = [preprocess.Message msg];     
elseif (preprocess.Evaluation == 4)
    msg = sprintf(' Testing on File %s, ', preprocess.input_file);
    preprocess.Message = [preprocess.Message msg];     
else 
    msg = sprintf(' Train-Test Split, Boundary: %d, ', preprocess.TrainTestSplitBoundary);
    preprocess.Message = [preprocess.Message msg]; 
end;

if (preprocess.MultiClassType == 0)
    msg = sprintf(' Classification, ', preprocess.TrainTestSplitBoundary);
    preprocess.Message = [preprocess.Message msg]; 
elseif (preprocess.MultiClassType == 1) 
    msg = sprintf(' Multiclass Classification Wrapper, ', preprocess.NumCrossFolder);
    preprocess.Message = [preprocess.Message msg];     
elseif (preprocess.MultiClassType == 2) 
    msg = sprintf(' Multilabel Classification Wrapper, ', preprocess.NumCrossFolder);
    preprocess.Message = [preprocess.Message msg];     
elseif (preprocess.MultiClassType == 3) 
    msg = sprintf(' Multiclass Active Learning Wrapper, ', preprocess.NumCrossFolder);
    preprocess.Message = [preprocess.Message msg];     
end;

if (preprocess.SVD == 1)
    msg = sprintf(' SVD ');
    preprocess.Message = [preprocess.Message msg]; 
end;

if (preprocess.Shuffled == 1)
    msg = sprintf(' Shuffled ');
    preprocess.Message = [preprocess.Message msg]; 
end;

if (preprocess.Sparse == 1)
    msg = sprintf(' Sparse ');
    preprocess.Message = [preprocess.Message msg]; 
end;

if (preprocess.MultiClass.CodeType == 0) 
    msg = sprintf(' Coding: 1-vs-r ');
    preprocess.Message = [preprocess.Message msg];
elseif (preprocess.MultiClass.CodeType == 1)
    msg = sprintf(' Coding: 1-vs-1 ');
    preprocess.Message = [preprocess.Message msg]; 
elseif (preprocess.MultiClass.CodeType == 2)
    msg = sprintf(' Coding: ECOC15_5 ');
    preprocess.Message = [preprocess.Message msg];
elseif (preprocess.MultiClass.CodeType == 3)
    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)