Archives: 3.1 release brings fast decision trees, hand-drawing classifiers and more
We are happy to announce immediate availability of perClass 3.1 release which introduces new fast decision tree and random forest classifiers, new interactive tools such as hand-drawn polygon classifier, hand-selection of thresholds, quick decisions and clustering in sdimage and much more. Enjoy!
PRSD Studio 3.0 release renamed to perClass
We are happy to announce immediate availability of 3.0 release which moves to the new name perClass. It brings significant update to the execution runtime, new powerful tools for handling images and extracting textural features and many enhancements and fixes. Hope you’ll enjoy it!
2.4 release with classifier execution from Excel worksheets
The latest 2.4.0 stable release brings classifier execution enhancements and examples. Classifiers may be now directly executed from Excel worksheets or directly on the command-line using the prsdrun utility.
Previous newletters available on-line
Our previous newsletters are now available from: http://perclass.com/index.php/newsletters/
Enjoy!
2.3 stable release available
Happy to announce availability of the 2.3.0 stable release enhancing sdcrossval command with per-fold execution speed measurements and fixing several issues. Enjoy!
2.2.5 release with simpler class subset selection, undo in scatter plot label painting and more
The 2.2.5 release brings support for regular expressions in class selection and renaming. This greatly simplifies definition of class subsets without tedious enumeration of names or indices. Interactive scatter now also supports regular expressions and provides number of user-requested productivity enhancements such as undo of last label painting operation, cycling through all classes in the current label set or class-to-top shortcut. New dissimilarity measures and minimum distance classifier, heavily used in spectral classification applications, were included. ROC plots may be zoomed. Enjoy!
Hosted Hypertools download
Download for Hypertools toolbox developed at TU Delft between 2003 and 2005 is now hosted at prsdstudio.com because it became inaccessible after the TU Delft web reorganization in mid-2010. Hypertools is currently not actively maintained or supported.
2.2.4 release brings linear and polynomial SVM, import from text files and Mac OS X 64-bit support
The latest 2.2.4 release brings support for linear and polynomial kernels in SVM with automatic grid search for parameters. Mac OS X 64-bit becomes the sixth supported platform. New sdimport and sdexport commands allow reading and writing of sddata sets from/to text files.
2.2.3 release brings identification of support vectors in training set and more
2.2.3 release allows to retrieve indices of support vectors in the training data set. Commands such as sddetector, sdfeatsel and sdsvc now support the ‘test’ option that may be used to specify an external test set leveraged in ROC optimization or parameter/grid search.
2.2.2 release: fast feature selection and visualization of decisions in high-D spaces
We are happy to announce new 2.2.2 release which brings fast feature selection scalable to large data sets and visualization of classifier decisions in higher-D spaces.
2.2.1 release: feature distributions in scatter plot and fast k-means clustering
First development release of the new 2.2 branch brings interactive feature distributions plots in the scatter (very useful to understand true nature of class overlap) and very fast k-means clustering which may be used as prototype selection for k-NN classifiers (1 million samples, 10 clusters in 3.3 sec). Enjoy!
new website design hosted at prsdstudio.com
The PR Sys Design website moved to a new design hosted at PRSDStudio.com domain. Please update your bookmarks. Hope you will like the new lightweight style. If you encounter any difficulties, please let us know at .
2.1 stable release brings multi-class SVM with automatic parameter selection
Happy to announce the immediate availability of the 2.1 stable release which brings RBF Support Vector Machine trained using the libSVM library. The sdsvc command automatically optimizes the sigma and C parameters and provides support for multi-class problems. Other new features include the sdreject command adding the reject option to any classifier (including multi-class discriminants) and many improvements to the cross-validation framework.
Enjoy!
new development release 2.0.8 brings interface simplification and stabilization
We are happy to announce the availability of new public development release 2.0.8. It cleans the user interface and documentation and fixes number of bugs. Detectors now automatically handle model output polarity (similarity for probabilistic models, distance for k-NN). There is now also a handy shortcut for displaying useful details on label objects such as class sizes and priors (lab’)
A new Getting started chapter in the documentation shows a step-by-step construction of a simple detector followed by a classifier.
Journal paper on multi-class ROC analysis in print
Happy to announce a our new paper to appear in Pattern Recognition Letters on construction of multi-class ROC. This work is a result of our cooperation with Thomas Landgrebe.
2.0 release brings new visualization and label handling!
After eight months of development and intensive testing with selected customers, we are happy to announce public availability of PRSD Studio 2.0!
This new release brings greatly enhanced visualization tools and improved handling of labels and data sets. New scatter plot allows you to limit the view to sample subsets based on label values. You can combine multiple filters and so investigate and understand your complex data. You may, for example, visualize only diseased tissue from selected patients using few clicks. You may view the data using different labels, define new classes, change markers directly in plots and enjoy much faster work with large data sets.
1.3 stable release is available
1.3 stable release is now available for all five platforms (Windows 32 and 64-bit, Linux 32 and 64-bit and Mac OS X)
classifier speedup by soft output approximation
We have just posted a new knowledge base article showing how to speed up accurate but slow non-parametric classifier such as Parzen or k-NN using neural network approximation of its soft outputs. This functionality is available in the latest 1.2.2 release of PRSD Studio.
1.2 stable release brings new classifiers
1.2 stable release brings new classifiers, namely the mixture of Gaussians automatically estimating number of components and the feed-forward neural network.
Next APR course scheduled for 26-30 October 2009
The next Advanced Pattern Recognition Course for Industry (APR) we are giving together with Delft Technical University (TU Delft) is scheduled for the week 26. - 30. October 2009. You can register here.
announcing the free version for non-commercial use: PRSD Studio Lite
We are happy to announce availability of the free version of PRSD Studio for non-commercial use: the PRSD Studio Lite. It is intended for people learning about pattern recognition or for evaluation purposes. PRSD Studio Lite is limited to data sets with maximum 300 samples and 3 classes.
Classifier embedded in pose recongition demo
Check out the video from pose recognition demo executing the final pose classifiers through libPRSD.
1.1.3 development version released
This development release brings some interesting features such as GUI for per-class visibility in sdscatter useful when handling large complex datasets. Support for operating points with reject option both close to and far from decision boundary. Leave-one-out over user-defined properties in a dataset (e.g. patient or object). It’s also fixing several bugs in sdscatter.
Next APR course scheduled for March
Coming Advanced pattern recognition course for industry (APR) is scheduled for 30.March. For details, see APR course web page.
New development release brings fast k-NN approximation
We have just release the 1.1.2 18-Nov-20008 development version bringing experimental version of a fast k-NN approximation for very large datasets and k-centres detector/discriminant. See example in our blog
Stable and development branches
PRSD Studio is now available in two branches—stable for the production use and the frequently-updated development one where new things happen. For details, see the announcement post in our Forum.
New PRTools forum open
The PRTools forum is now hosted here, at PR Sys Design! If you have questions or remarks about PRTools, you are welcome to join the discussion.
New release with interactive scatter and image viewer
We’ve just released PRSD Studio 1.1 with number of enhancements. The most interesting are perhaps the user-friendly interactive scatter plot and image viewer allowing for hand-painting sample/pixel labels and renaming the classes.
First release available for download
1.0 release is here! You can download Windows and Mac evaluation version right now. Enjoy :-)
Compiled on embedded platform
We’re happy to announce that libPRSD was successfully compiled and tested on embedded platform (TI C64x DSP). This extends our offering from the PC world to high-speed industrial devices. In case of interest please contact us for details.
Linux version now available
There is now Linux x86 build available! It brings also new features, such as decision tree support.
New knowledge base tips available
There are some interesting new How to... tips in our Knowledge base!
October APR Course
Are you working on practical classification problems but you miss the necessary knowledge or skills concerning the use of good statistical techniques to analyze and learn from your data?
The Advanced Pattern Recognition course offers you a unique opportunity to improve your knowledge of pattern recognition. We organize this course in collaboration with Delft Technical University. The next course is scheduled on 27-31 October 2008 . Registrer now.
New site design
We have updated the PRSD website and included new support forum.
APR course in June
Are you working on practical classification problems but you miss the necessary knowledge or skills concerning the use of good statistical techniques to analyze and learn from your data?
The Advanced Pattern Recognition course offers you a unique opportunity to improve your knowledge of pattern recognition. We organize this course in collaboration with Delft Technical University. The next course is scheduled on 9-13 June 2008 .