What can be expected

What can be expected

The target of the PRTools project is to give students and researchers a consistent set of Matlab commands for creating pattern recognition systems in the domain of sensory data (e.g. images, time signals, spectra). The toolbox thereby offers routines for:

• the analysis of raw data, organized in directories on disk.
• the representation in vector spaces.
• the generalization of datasets in vector spaces by training classifiers.
• the evaluation of classifiers.

Some limitations of the toolbox are:

• the toolbox is not well suited for symbolic data, graphs, strings or for analyzing nominal data.
• the handling of missing values is prepared but there are (almost) no supporting routines implemented.
• there is just a basic set of routines for cluster analysis.
• routines are not written with an emphasis on minimizing computational complexity.
• almost all routines are written in Matlab, causing overhead in the memory demand and slowing down the computational speed.

Users can extend the toolbox with their own routines, or build interfaces with other Matlab packages. There are interface routines with LibSVM, C4.5 and DIPLib available. For popular public domain datasets PRTools versions can de downloaded. It is simple to load datasets in ASCII format or images in one of the Matlab readable formats.

What can be expected