PRSD Studio features

Flip through features, analyze multiple sample groupings (labels, object id, material types,...) or define new sample groupings

Define new classes, name plausible clusters and identify outliers by painting in scatter plots

Enhance the scatter plots with backdrop of the algorithm decisions or soft outputs in 2D feature spaces

In adition to standard two-class ROC analysis, optimize classifier operating points in multi-class situations.

Allows for designing detectors (distance-based rejection, one-class classification)

For two-class or multi-class discrimination.

Class errors, TPr, FPr, TNr, FNr, precision, specificity, sensitivity, positive fraction, any field of confusion matrix.

Define custom performance measures based on confusion matrix.

Get the raw data out and perform customized cost-based optimization or classification with constraints.

Variance information allows for testing of statistical significance regarding performance differences between different operating points on the same ROC. General operating point averaging approach is adopted applicable to two-class, multi-class, arbitrary operating point definition and arbitrary ROC optimizer.

Inspect and select operating points interactively in ROC spaces. Flip between multiple performance measures.

Use stratified N-fold rotation or randomization cross-validation to estimate performance of your algorithms.

Fix the random seed so that the experiments are repeatable.

Support for cross-validation keeping multiple replicas (measurements) of one sample either in the training set or in the test set.

Run cross-validation at a set of operating points i.e. estimate ROC with variances at apriori fixed operating points.

Estimate unbiased classifier outputs using cross-validation with one command. Practical technique for definition of operating points in ROC analysis or training multi-stage systems such as trainable combiners.

Inspect the training and test sets for a specific fold and per-fold trained algorithms.

Cross-validate complex systems such as detector-classifier cascades where each stage was already ROC optimized and the whole system only returns crisp decisions.

Growing from our own consultancy projects, we implemented a set of practical algorithms useful in real-world situations.

Execute your trained algorithms in your deployment application faster than they were running in your Matlab prototype.

Bring your trained algorithms into deployment environment in minutes, not weeks.

Changing the algorithm in your production application becomes the matter of a simple save and load. Ship your applications with many classifiers your customers may choose from.

The library is implemented in ANSI C and fully self-contained.

The libPRSD interface is clean and simple. Initialize the library, load the pipeline, connect your data buffers to it and execute. Why to make it more complicated?

Embedding the library to your application is very simple and usually takes about an hour.

Identical code, pipeline format and interface across all platforms.

No more hunting for parameters of your classifiers again! Cannonical parameters of all algorithms implemented in libPRSD are accessible through the pipeline object in Matlab.

Pipeline files are encrypted with the industry standard encryption technology to protect intellectual property in your classifiers.

Running application may switch between several operating points. This allows the operator to adjust the system to current conditions (population 'screening' mode and 'cancer patient analysis' mode)

Use PRSD Studio at any of your machines with a hardware dongle.

Easily describe your specific pattern recognition algorithms starting from measurements and finishing with crisp decisions.

Define parameters so one function can serve as a blueprint for the whole family of algorithms varying models, smoothing parameters, numers of components etc.

Custom algorithm may be saved and used later each preserving all information needed to perform decisions including fine-tuned operating points.

PRSD Toolbox is seamlessly integrated with PRTools, the leading academic toolbox for pattern recognition.