Classer is a set of software tools for applying machine learning classifier models to arbitrary data sets. Layered on top of implementations of ARTMAP neural networks, the Classer toolkit lets the user define classifier models, apply them to process data sets, and automate output data collection and parameter space exploration.
ClasserScript is a batch interface to the Classer toolkit. More concretely, ClasserScript is a miniature language for defining views of data, classifier models, and iterative combinations of the two to generate a variety of output metrics.
In addition to classifier simulation management, version 1.1 offers a set of tools implementing the ARTMAP Fusion System for rule discovery, a process for extracting knowledge from a data set in the form of rules relating data classes to each other.
For users needing lower-level access to the ARTMAP implementations underlying Classer, the ARTMAP code is available separately, both as source code, and as a library (Windows DLL). The entry-points to the ARTMAP code are described by HTML API documentation. API documentation for the larger Classer project is also available here, along with a use cases document describing how to use the API.
As a simple illustration of the functionality of ARTMAP networks, a Java Applet is provided. The demo is based on a Java port of Classer's ARTMAP code, which is also available for download.
Finally, many of the examples that are provided use one of three data sets, which are provided for download here.
The Classer toolkit has been developed by Siegfried Martens, a post-doctoral fellow in the Department of Cognitive and Neural Systems at Boston University.
Decision surfaces from Classer showing generalization of the Letters data set
Visualizations of the Boston Testbed using Classer
Carpenter, G.A., Martens, S., & Ogas, O.J. (2005). Self-organizing information fusion and hierarchical knowledge discovery: a new framework using ARTMAP neural networks. Neural Networks, 18(3) 287-295. Technical Report CAS/CNS TR-2004-016, Boston, MA: Boston University.
Carpenter, G.A. (2003).“Default ARTMAP”, Proceedings of the International Joint Conference on Neural Networks (IJCNN'03), Portland, Oregon, 1396-1401
This software is provided free of charge. As such, the author assumes no responsibility for the program’s behavior. While they have been tested and used in-house for a year, no claim is made that Classer, ClasserScript, or the encapsulated ARTMAP classifier implementations are correct or bug-free. They are used and provided solely for research and educational purposes. No liability, financial or otherwise is assumed regarding any application of ClasserScript.
The development of Classer and ClasserScript would not have been possible without the support of Dr. Gail Carpenter or of the Technology Laboratory at Boston University’s Department of Cognitive and Neural Systems. The author was supported in this work by postdoctoral fellowships from the National Geospatial-Intelligence Agency and the National Science Foundation (NMA 501-03-1-2030 and NSF DGE-0221680);
The Frey and Slate Letters data set was obtained from the UCI Repository (http://www.ics.uci.edu/~mlearn/MLRepository.html).
The Boston testbed was obtained in part thanks to BU’s Department of Remote Sensing, and Suhas Chelian and Brad Rhodes helped create the ground truth set.
Gail Carpenter and others developed the ARTMAP family of neural networks that form Classer’s computational core.
Have questions? Found a bug? Please contact us.