PRTools Updates (Release Notes)

8 March 2017

Version 5.3.3 minor updates

  • beautifications, minor bug fixes
  • adaptations to Matlab’s changed graphic systems
  • more easy use of batch processing by prbatch
  • hierarchical clustering by hclust is extended and computes faster
  • copulam added for nonlinear mappings
  • mapsd added for mapping new objects to given non-linear representations

31 August 2016

Version 5.3.2 minor updates

2 September 2015

Version 5.3.1 minor updates

  • Extended call to the SVM classifier in the statistical toolbox by statssvc.
  • Some changes in baggingc and rsscc in relation to combining.
  • Other beautifications.

16 August 2015

Version 5.3 control of computing time and handling of degenerate training sets

  • Classifiers can now be trained with degenerate training sets having missing or very small classes.
  • Various optimizers as used in neural networks, non-linear mappings and the generation of base classifiers like in adaboostc can be stopped by a user adjustable maximum computing time setting, see prtime.
  • The warning system as controlled by prwarning has been redefined.
  • An interface to the Support Vector classifier (svmtrain) available in Matlab’s Stats toolbox has been added by routines for linear (statssvc) and radial basis (pkstatssvc and rbstatssvc) classifiers.
  • Some data generation routines are added: gendatmm (a multi-modal version of gendatm) and genmdat (multi-dimensional versions of existing two-dimensional datasets).

21 December 2014

Version 5.2.3 minor updates and bug fixes

  • Some problems have been solved related to the new graphical software of Matlab 2014b.
  • A new routine for prototype selection: selproto
  • Other bugfixes and minor changes

13 September 2014

Version 5.2.1 Better tools for categorical features and visualization

  • The creation and handling of datasets with categorical features: cat2dset, cat2real and cat2feat. Categorical and integer features can be used for creating dataset labels by feat2lab.
  • The conversion of cell arrays to datasets and the other way around: cell2dset and dset2cell.
  • A new routine for dataset visualization: tsnem. An example file prex_mds compares various methods.
  • Improved methods for handling missing values: misval and ismisval.
  • A faster routine for computing histograms from image datasets by given numbers of bins: im_hist.
  • A simplified scatterplot: scattern.

14 May 2014

Version 5.1.1, resolving some naming conflicts.

The naming conflicts of the PRTools routines kmeans, roc and plotsom have been solved in such a way that they can still be called for variables of the types prdataset and prmapping, but do not shade other routines with the same name.

  • PRToolskmeans has been renamed into prkmeans. A new routine kmeans has been created in the @prdataset subdirectory. It calls prkmeans in the PRTools main directory. The kmeans routine in the Mathwork’s stats toolbox is now available for dealing with data given by doubles.
  • PRToolsroc has been renamed into prroc. A new routine roc has been created in the @prdataset subdirectory. It calls prroc in the PRTools main directory. The roc routine in the Mathwork’s nnet toolbox is now available for dealing with data given by doubles.
  • PRToolsplotsom has been renamed into prplotsom . A new routine plotsom has been created in the @prmapping subdirectory. It calls prplotsom in the PRTools main directory. In this way it will not be confused with the obsolete plotsom routine in the nnet toolbox..

April 2014

Version 5.1, beautifications, bug fixes and small upgrades.

28 August 2013

Version 5.0.2, bug fixes, small upgrades

15 July 2013

A new version, PRTools5, is now available.
It has been created to make PRTools compatible with the Stats Toolbox of Matlab.

  • Version 5, similar to version 4 but not fully compatible.
  • The classes dataset, datafile and mapping are renamed into prdataset, prdatafile and prmapping
  • The routines map, crossval and pca are renamed into prmap, prcrossval and pcam
  • prload can be used to load old mat-files
  • prtools4to5 can be used to convert PRTools4 routines to PRTools5
  • Matlab Stats Toolbox classifiers are integrated as statsdtc, statsknnc, statslins and statsnbc.

There is a page with more information.

8 July 2013

  • Version 4.2.5, bug fixes and beautifications
  • concatm: mapping concatenating datasets or mappings
  • cat2data: adding nominal features to a dataset
  • feat2lab: use a feature to label a dataset
  • gendatv: generation of a very large, multiclass dataset
  • findclasses: retrieve class wise indices of objects in a dataset
  • faster handling of datasets having many (thousands of) classes
  • confmat: may return objects corresponding to the elements of the confusion matrix
  • mapex: Train and execute untrained mapping in one flow
  • mapm: mapping any Matlab command
  • mappingtools: import of a set of mapping tools

January 25, 2013

There is a post discussing updates to this point

November 26, 2012

  • Version 4.2.3, just minor bugs fixes and beautifications
  • PRTools is now distributed by 37 Steps

September 15, 2012

  • Version 4.2.2
  • New classifier: dtc, Verzakov Tree -Decision Tree Classifier
  • New classifier: randomforestc, Breiman’s Random Forest
  • Programming tools added enabling a more simple definition of mappings and classifiers: setdefaultsmapping_tasktrained_mappingtrained_classifier. For an example see randomforestc
  • Routine added for changing PRTools global variables: prglobal
  • Handling of doubles in mappings: where possible mappings will accept arrays of doubles instead of datasets without internal conversion to a dataset. In such cases outputs will be an array of doubles as well
  • New routine for class selectionselclass. This routine is especially useful for multi-labeled data as it may select a class defined by another label list than the actual one. For such usage it is much more user friendly than seldat

September 3, 2011

  • version 4.2.1, this version is identical to 4.2.0, except that p-codes are generated by Matlab 7.11 instead of Matlab 6.5.
    Please inform us when old p-codes are still needed.

August 26, 2011

  • version 4.2.0 (wrong Contents file, later corrected)
  • better, extended, soft label implementation
  • ‘bag of objects’ classification (bagc)
  • some missing value handling (misval)
  • standard PRTools datasets are downloaded when needed (prdatasets)
  • standard PRTools datafiles are downloaded when needed (prdatafiles)
  • minor changes and bug fixes

December 12, 2010

  • version 4.1.12
  • many minor bugs are solved

June 25, 2010

  • version 4.1.10
  • option for density preserving data splitting added to crossval, thanks to Budka and Gabrys
  • new classifier drbmc: Discriminative Restricted Boltzmann Machine, thanks to Laurens van der Maaten
  • small changes in traincc and wvotec

June 1, 2010

  • version 4.1.9
  • bug in wvotec fixed
  • libsvc and nulibsvc added (expects libsvm package)

May 3, 2010

  • bug in indexing datasets fixed

April 2010

  • vpc, voting perceprton classifier implemented by Laurens van der Maaten.
  • prarff, import of WEKA ARFF files available.

March 2010

  • communication with users improved by the command prtools, generating a web interface.

February 2010

  • prwaitbar usage simplified by implementation of prwaitbarinit, prwaitbaronce and prwaitbarnext

July 2009

  • featselv implemented

March 2009

  • dcsc (dynamic classifier selection) and modelsc added as combining classifiers.

January 2009

  • naivebcc and mlrc: new trainable combiners implemented by Chunxia Zhang.
  • loso, leave-one-set out crossvalidation implemented.
  • testc now accepts missing classes in case of undefined (empty) priors.
  • im_patch for generation of image patches implemented in datafiles
  • band2obj and bandsel for handling image bands implemented in datafiles
  • 1-d images (time signals) are now accepted for datafiles, patches and bands

August 2008

  • im_dbr constructed: GUI to illustrate PRTools usage for image database retrieval.

July 2008

  • prwaitbar implemented and included in many routines.
  • ident field in dataset reorganised to avoid overhead.
  • createdatafile implemented to support saving partially constructed datafiles.

June 2008

  • adaboostc added

August 2007

  • A set of regression routines added by David Tax: linearr, ridgerlassor, svmrksmoothrknnrpinvrplsrplsmtestrrsquared

June 2007

  • performance measures for testc extended.
  • automatic parameter optimisation implemented for many routines by regoptc.
  • rbsvc, automatic radial basis support  vector classifier, implemented.
  • rsscc, random subspace combining classifier implemented.

May 2007

  • naivebc extended: allows now also densitites estimated by parzenm and gaussm.
  • soft output nmc improved: now based and spherical Gaussian densities.
  • many image preprocessing routines implemented for datafiles, e.g. im_bdilation, im_gaussf,im_minf,im_skel, im_berosion, im_gray,im_moments, im_skel_meas, im_box,im_harris,im_norm,im_stat, im_bpropagation, im_hist_equalize, im_stretch, im_center, im_invert, im_profile, im_threshold, im_label, im_resize, im_unif, im_fft,im_maxf, im_rotate, im_fill_norm, im_mean, im_scale, im_gauss, im_measure, im_select_blob.

April 2007

  • normal_map (used in ldc,qdc,udc) now stores inverted covariances: slower training, faster execution.
  • gaussm and mogc improved: increased stability by adding noise, more trials, better regularisation.

March 2007

  • Datafiles ready for usuage. Consequence is that as feature sizes of datafiles cannot be set appropriately, also mappings should allow for unset feature size (featsize = 0). Still much debugging to be done.

December 2006

  • PRMEMORY, PRTRACE, GRIDSIZE, etc are made persistent instead of global
  • plotr renamed into plote.
  • general kernel option implemented for svc and nusvc

November 2006

  • testauc (area under the curve) added.
  • testc upgraded.

September 2006

  • implementation of the multiple labeling, see multi-labeling.
  • structure implementation for user and ident field of dataset.

March 2006

  • intial implementation of the datafile class.

October 2005

  • combining superfluous prwarning calls.
  • bhatm, Bhattacharryya feature reduction implemented

April 2005

  • labcmp added, useful for comparing label sets.

January 2005

  • change of policy for prwarning levels: by default just print level 1 warnings.
  • upgrade warning level for use of class frequencies in case of empty priors.

December 2004

  • parzendc, parzen_map, nmc, allowing soft labels
  • plotr (later plote), allow cells of error structures

October 2004

  • nusvc_nu added

September 2004

  • gaussm and emclust: get right number of components
  • gaussm and emclust: improved soft label handling
  • plotr (later plote): mulitple error curves implemented
  • scalem, variance option corrected

August 2004

  • output of confmat and testc upgraded by disperror.
  • proxm extended.

June 2004

  • fdsc, feature based dissimilarity space classifier implemented.

May 2004

  • ploto, one-dimensional plot of features, added.
  • immoments, conversion of object images to moments, added.
  • crossvalidation in feature selection routines implemented.