% PRTools3.0 December 1999% % PRTOOLS is a basic set of statistical pattern recognition% tools under Matlab. Not all commands are entirely tested.%% It is heavily upgraded from the previous version, November 1997%% The main change is the use of "classes' and 'objects' offered% by Matlab-5 (Don't confuse them with the pattern recognition% classes and objects!). This simplifies the use, but may be% hard to understand for old users. Moreover, it makes PRTools3.0% completely incompatible with old versions. %% There are two of these class-constructors: datasets and mappings,% the basic elements of pattern recognition. As use is made of% structures (entirely hidden for the user), they contain all types% of information that should be transferred otherwise explicitely% by the user: labels, sizes, weigths, mapping types, etcetera.%% What we can do now are defining datasets by%   A = dataset(a,labels);% use it for training by defining a mapping by%   W = ldc(A);% which may also be written as%   W = A*ldc;% and use it for testing by:%   e = A*W*testd;% This can be read as: map the dataset A on W (i.e. a space constructed% by class memberships) and test it (count the errors).% Mappings can be combined sequentially (W = W1*W2), by stacking% ([W = [W1 W2 W3]) and in parallel (W = [W1; W2; W3]). This facilitates% the construction of combined classifiers, e.q. W = maxc([W1 W2 W3]).%% A number of routines has been renamed, e.g.:%   nlc     : ldc%   nqc     : qdc%   mlslc   : fisherc%   nmlc    : nmc% Others are combined or have been moved outside the user sight into the% private directory. Many routines have been added. See the Contents file.%% PRTools3.1, January 2000%% For affine mapping and normalizing mappings the multiplicative output% constant w.v is now integrated in the mapping coefficients. w.v is set% to one.%% Untrained mappings are no longer empty, so now isempty(fisherc) is 0.% These mappings can now be detected by the new routine 'istrained', so% istrained(fisherc) is 0.%% The display command for datasets has been changed such that it returns% the actual number of classes available in the dataset.%% New commands:%	seldat:   Select classes / features / objects from dataset%   classim:  Classify image using a given classifier%   clevalf:  Classifier evaluation (feature size curve)%   cnormc:   Renamed copy of normc%   emclust:  Expectation - Maximization Clustering%   spatm:    Augment image dataset with spatial label information%   nlfisherm:Renamed copy of nlklm%% PRTools3.1.2, January 2001%% The dataset structure changed in order to speed up the handling of labels.% This should not effect the calls as it is intended to be upwards compatible.% Newly created datasets, however, cannot be used under previous versions% of PRTools.%% The confmat command changed slightly to enable a more general use. As a% result the number of errors for the specific useage in which the two% labellists are identical is not returned anymore.%% A new command matchlab is added to rotate labels for optimal match.%% PRTools3.1.3, July 2001%% New commands%   pca:      Principal component analysis (replaces overloaded procedure by klm)%   maf:      Maximum autocorrelation mapping, (pca for multi-band images)%% Consequently, some of the old, overloaded and confusing possiblities in klm% are removed.%% PRTools3.1.4, August 2001%%   some bugs are removed in @dataset/subsasgn.m and @dataset/dataset.m%   all relating to the erroneous working of expressions like%   A([1 2 3],:) = []; A(:,[1 2 3]);%   if A is a dataset.%%   kljlc:   bug removed%   gauss:   label generation improved%   gendats: labels changed from numeric to character%   featsel* redundant output suppressed%% PRTools3.1.5, August 2001%%   knnc:    Makes no use anymore of prior probabilities.%% New command%   getprob: Retrieves dataset prior probabilities%% PRTools3.1.6, September 2001%%   normal_map: now generates scaled densities%   qdc:        now also in 2-class problems based on densities%   plotm:      internal scaling improved. Parameter added for%               selecting contour.%   scatterd:   extended, a.o. with 3d plot%   distm:      now returns a dataset%   gendat:     relative class sizes supported%% New commands:%%   subsc:    Subspace Classifier%   quadrc:   Quadratic classifier (original 2-class qdc)%   mclassc:  Multi-class classifier by 2-class discriminants%   classsizes: Returns sizes of classes in a dataset%   getlablist: Returns label list of a dataset%               See 'help datasets' for all means to retrieve data stored%               in a dataset.%   mds:      Non-linear mapping by multi-dimensional scaling (Sammon)%   classs:   Multi-dimensional mapping by classical scaling%% *************************************************************%% More information can be found by%   help prtools%   help datasets%   help mappings% or in the manual (http://www.ph.tn.tudelft.nl/~bob/postscript/PRTools3.ps)%% This software can be used freely for inspection and academic research.% Please refer to%%   R.P.W. Duin, PRTools 3, A Matlab Toolbox for Pattern Recognition, Delft%   University of Technology, January 2000%% if it has been useful to you. If you like to use the toolbox for commercial% purposes, please contact me.%% Bob Duin%% Copyright: R.P.W. Duin, duin@ph.tn.tudelft.nl% Faculty of Applied Physics, Delft University of Technology% P.O. Box 5046, 2600 GA Delft, The Netherlands