Tuesday, February 16, 2010
Friday, February 12, 2010
Bayesian Network Repository
--------------------------------------------------------------------------------
Contents
About This Page
The datasets
Network formats and Utilities
Related Sites
--------------------------------------------------------------------------------
Mission
Our in intention is to construct a repository that will allow us empirical research within our community by facilitating (1)better reproducibility of results, and (2) better comparisons among competing approach. Both of these are required to measure progress on problems that are commonly agreed upon, such as inference and learning.
A motivation for this repository is outlined in "Challenge: Where is the impact of Bayesian networks in learning?" by N. Friedman, M. Goldszmidt, D. Heckerman, and S. Russell (IJCAI-97).
This will be achieved by several progressive steps:
Sharing domains. This would allow for reproduction of results, and also allow researchers in the community to run large scale empirical tests.
Sharing task specification. Sharing domains is not enough to compare algorithms. Thus, even if two papers examine inference in particular network, they might be answering different queries or assuming different evidence sets. The intent here is to store specific tasks. For example, in inference this might be a specific series of observations/queries. In learning, this might be a particular collection of training sets that have a particular pattern of missing data.
Sharing task evaluation. Even if two researchers examine the same task, they might use different measures to evaluate their algorithms. By sharing evaluation methods, we hope to allow for an objective comparison. In some cases such evaluation methods can be shared programs, such as a program the evaluates the quality of learned model by computing KL divergence to the "real" distribution. In other cases, such an evaluation method might be an agreed upon evaluation of performance, such as space requirements, number of floating point operations, etc.
Organized competitions. One of the dangers of empirical research is that the methods examined become overly tuned to specific evaluation domains. To avoid that danger, it is necessary to use "fresh" problems. The intention is to organize competitions that would address a specific problems, such as causal discovery, on unseen domains.
Plans for the future
Currently, this site contains several domains. The plan is to gradually add other components discussed above.
Please send suggestions and contributions to galel@cs.huji.ac.il.
Acknowledgements
Thanks to Fabio Cozman, Bruce D'Ambrosio, Moises Goldszmidt, David Heckerman, Othar Hansson, Daphne Koller, and Stuart Russell for discussions about the organization of this site. Thanks to John Binder, Jack Breese, David Heckerman, Uffe Kjaeruff, and Mark Peot, for contributing networks.
--------------------------------------------------------------------------------
galel@cs.huji.ac.il
Contents
About This Page
The datasets
Network formats and Utilities
Related Sites
--------------------------------------------------------------------------------
Mission
Our in intention is to construct a repository that will allow us empirical research within our community by facilitating (1)better reproducibility of results, and (2) better comparisons among competing approach. Both of these are required to measure progress on problems that are commonly agreed upon, such as inference and learning.
A motivation for this repository is outlined in "Challenge: Where is the impact of Bayesian networks in learning?" by N. Friedman, M. Goldszmidt, D. Heckerman, and S. Russell (IJCAI-97).
This will be achieved by several progressive steps:
Sharing domains. This would allow for reproduction of results, and also allow researchers in the community to run large scale empirical tests.
Sharing task specification. Sharing domains is not enough to compare algorithms. Thus, even if two papers examine inference in particular network, they might be answering different queries or assuming different evidence sets. The intent here is to store specific tasks. For example, in inference this might be a specific series of observations/queries. In learning, this might be a particular collection of training sets that have a particular pattern of missing data.
Sharing task evaluation. Even if two researchers examine the same task, they might use different measures to evaluate their algorithms. By sharing evaluation methods, we hope to allow for an objective comparison. In some cases such evaluation methods can be shared programs, such as a program the evaluates the quality of learned model by computing KL divergence to the "real" distribution. In other cases, such an evaluation method might be an agreed upon evaluation of performance, such as space requirements, number of floating point operations, etc.
Organized competitions. One of the dangers of empirical research is that the methods examined become overly tuned to specific evaluation domains. To avoid that danger, it is necessary to use "fresh" problems. The intention is to organize competitions that would address a specific problems, such as causal discovery, on unseen domains.
Plans for the future
Currently, this site contains several domains. The plan is to gradually add other components discussed above.
Please send suggestions and contributions to galel@cs.huji.ac.il.
Acknowledgements
Thanks to Fabio Cozman, Bruce D'Ambrosio, Moises Goldszmidt, David Heckerman, Othar Hansson, Daphne Koller, and Stuart Russell for discussions about the organization of this site. Thanks to John Binder, Jack Breese, David Heckerman, Uffe Kjaeruff, and Mark Peot, for contributing networks.
--------------------------------------------------------------------------------
galel@cs.huji.ac.il
Graphical Models -software tools
Working Group Neural Networks and Fuzzy Systems
Graphical Models
Software Tools back to the main page
Contents
Overview
BayesBuilder
Bayesian Knowledge Discoverer / Bayesware Discoverer
Bayes Net Toolbox
Belief Network Power Constructor
GeNIe / SMILE
Hugin
Netica
Pulcinella
Tetrad
WinMine / MSBN
Overview
On this page we briefly describe some software tools that support reasoning with graphical models and/or inducing them from a database of sample cases. Of course, we do not claim this list to be complete (definitely it is not). Nor does it represent a ranking of the tools, since they are ordered alphabetically. More extensive lists of probabilistic network tools have been compiled by
Russel Almond (an old list, which is not maintained anymore):
http://www.stat.washington.edu/almond/belief.html
Kevin Patrick Murphy:
http://www.cs.berkeley.edu/~murphyk/Bayes/bnsoft.html
and Google:
http://directory.google.com/Top/Computers/Artificial_Intelligence/Belief_Networks/Software/
The Bayesian Network Repository is also a valuable resource. It lists examples of Bayesian networks and datasets, from which they can be learned:
http://www.cs.huji.ac.il/labs/compbio/Repository/
The software we developed in connection with our book is available at:
http://fuzzy.cs.uni-magdeburg.de/books/gm/software.html
Tools for troubleshooting Microsoft products, which are based on Bayesian networks (but do not allow you to access them directly), can be found at
http://support.microsoft.com/support/tshoot/default.asp
back to the top
BayesBuilder
SNN, University of Nijmegen
PO Box 9101, 6500 HB Nijmegen, The Netherlands
http://www.mbfys.kun.nl/snn/Research/bayesbuilder/
BayesBuilder is a tool for (manually) constructing Bayesian networks and drawing inferences with them. It supports neither parameter nor structure learning of Bayesian networks. The graphical user interface of this program is written in Java and is easy to use. However, the program is available only for Windows, because the underlying inference engine is written in C++ and has only been compiled for Windows yet. BayesBuilder is free software.
back to the top
Bayesian Knowledge Discoverer / Bayesware Discoverer
Knowledge Media Institute / Department of Statistics
The Open University
Walton Hall, Milton Keynes MK7 6AA, United Kingdom
http://kmi.open.ac.uk/projects/bkd/
Bayesware Ltd.
http://bayesware.com/
The Bayesian Knowledge Discoverer is a software tool that can learn Bayesian networks from data (structure as well as parameters). The dataset to learn from may contain missing values, which are handled by an approach called "bound and collapse" that is based on probability intervals. The Bayesian Knowledge Discoverer is free software, but it has been succeeded by a commercial version, the Bayesware Discoverer. This program has a nice graphical user interface with some powerful visualization options. A 30 days trial version may be retrieved free of charge. Bayesware Discoverer is available for Windows, Unix and Macintosh.
back to the top
Bayes Net Toolbox
Kevin Patrick Murphy
Department of Computer Science, UC Berkeley
387 Soda Hall, Berkeley, CA 94720-1776, USA
http://www.cs.berkeley.edu/~murphyk/Bayes/bnt.html
The Bayes Net Toolbox is an extension for Matlab, a well-known and widely used mathematical software package. It supports several different algorithms for drawing inferences in Bayesian networks as well as several algorithms for learning the parameters and the structure of Bayesian networks from a dataset of sample cases. It does not have a graphical user interface of its own, but profits from the visualization capabilities of Matlab. The Bayes Net Toolbox is distributed under the Gnu Library General Public License and is available for all systems that can run Matlab, an installation of which is required.
back to the top
Belief Network Power Constructor
Jie Cheng
Dept. of Computing Science, University of Alberta
155 Athabasca Hall, Edmonton, Alberta, Canada T6G 2E1
http://www.cs.ualberta.ca/~jcheng/bnpc.htm
The Bayesian Network Power Constructor uses a three phase algorithm that is based on conditional independence tests to learn the structure of a Bayesian network from data. The conditional independence tests rely on mutual information, which is used to determine whether a (set of) node(s) can reduce or even block the information flow from one node to another. The program comes with a graphical user interface, though a much less advanced one than those of, for instance, HUGIN and Netica (see below). It does not support drawing inferences, but has the advantage that it is free software. It is available only for Windows.
back to the top
GeNIe / SMILE
Decision Systems Laboratory, University of Pittsburgh
B212 SLIS Building, 135 North Bellefield Avenue, Pittsburgh, PA 15260, USA
http://www2.sis.pitt.edu/~genie/
SMILE (Structural Modeling, Inference and Learning Engine) is a library of functions for building Bayesian networks and drawing inferences with them. It does support neither parameter nor structural learning of Bayesian networks. GeNIe (Graphical Network Interface) is a graphical user interface for SMILE, that makes the functions of SMILE easily accessible. While SMILE is platform independent, GeNIe is available only for Windows, since it relies heavily on the Microsoft Foundation classes. Both packages are distributed free of charge.
back to the top
Hugin
Hugin Expert A/S
Niels Jernes Vej 10, 9220 Aalborg, Denmark
http://www.hugin.com
Hugin is one of the oldest and best-known tools for Bayesian network construction and inference. It comes with an easy to use graphical user interface, but also has an API (application programmers interface) for several programming languages, so that the inference engine can be used in other programs. It supports estimating the parameters of a Bayesian network from a dataset of sample cases. In a recent version it has also been extended by a learning algorithm for the structure of a Bayesian network, which is based on conditional independence tests. Hugin is a commercial tool, but a demo version with restricted capabilities may be retrieved free of charge. Hugin is available for Windows and Solaris (Sun Unix).
back to the top
Netica
Norsys Software Corp.
2315 Dunbar Street, Vancouver, BC, Canada V6R 3N1
http://www.norsys.com
Like Hugin, Netica is a commercial tool with an advanced graphical user interface. It supports Bayesian network construction and inference and also comprises an API (application programmers interface) for C++, so that the inference engine may be used in other programs. Netica offers quantitative network learning (known structure, parameter estimation) from a dataset of sample cases, which may contain missing values. It does not support structural learning. A version of Netica with restricted capabilities may be retrieved free of charge, but the price of a full version is also moderate. Netica is available for Windows and Macintosh.
back to the top
Pulcinella
IRIDA, Université Libre de Bruxelles
50, Av. F. Roosevelt, CP 194/6, B-1050 Brussels, Belgium
http://iridia.ulb.ac.be/pulcinella/Welcome.html
Pulcinella is more general than the other programs listed on this page, as it is based on the framework of valuation systems [Shenoy 1992a]. Pulcinella supports reasoning by propagating uncertainty with local computations w.r.t. different uncertainty calculi, but does not support learning graphical models from a dataset of sample cases in any way. The current version of Pulcinella does not have a graphical user interface, but an outdated version of such an interface may be retrieved for Solaris (Sun Unix). Pulcinella is available for Solaris (Sun Unix) and Macintosh, but requires a Common Lisp system.
back to the top
Tetrad
Tetrad Project, Department of Philosophy
Carnegie Mellon University, Pittsburgh, PA, USA
http://hss.cmu.edu/html/departments/philosophy/TETRAD/tetrad.htm
Tetrad is based on the algorithms developed in [Spirtes et al 1993], i.e. on conditional independence test approaches to learn Bayesian networks from data, and, of course, subsequent research in this direction. It can learn the structure as well as the parameters of a Bayesian network from a dataset of sample cases, but does not support drawing inferences. Currently the program is being ported to Java (Tetrad IV). Older versions are available for MSDOS (Tetrad II) and Windows (Tetrad III). Tetrad II is commercial, but available at a moderate fee. Free beta versions are available of Tetrad III and Tetrad IV.
back to the top
WinMine / MSBN
Machine Learning and Statistics Group
Microsoft Research, One Microsoft Way, Redmond, WA 98052-6399, USA
http://research.microsoft.com/~dmax/WinMine/tooldoc.htm
WinMine is a toolkit, i.e. a set of programs for different tasks, rather than an integrated program. Most programs in this toolkit are command line driven, but there is a graphical user interface for the data converter and a network visualization program. WinMine learns the structure and the parameters of Bayesian networks from data and uses decision trees to represent the conditional distributions. It does not support drawing inferences. However, Microsoft Research also offers MSBN (Microsoft Bayesian Networks), a tool for (manually) building Bayesian networks and drawing inferences with them, MSBN comes with a graphical user interface. Both programs, WinMine as well as MSBN, are available for Windows only.
back to the top
© 2002 Christian Borgelt
Last modified: Fri Oct 25 11:05:52 MEST 2002
Graphical Models
Software Tools back to the main page
Contents
Overview
BayesBuilder
Bayesian Knowledge Discoverer / Bayesware Discoverer
Bayes Net Toolbox
Belief Network Power Constructor
GeNIe / SMILE
Hugin
Netica
Pulcinella
Tetrad
WinMine / MSBN
Overview
On this page we briefly describe some software tools that support reasoning with graphical models and/or inducing them from a database of sample cases. Of course, we do not claim this list to be complete (definitely it is not). Nor does it represent a ranking of the tools, since they are ordered alphabetically. More extensive lists of probabilistic network tools have been compiled by
Russel Almond (an old list, which is not maintained anymore):
http://www.stat.washington.edu/almond/belief.html
Kevin Patrick Murphy:
http://www.cs.berkeley.edu/~murphyk/Bayes/bnsoft.html
and Google:
http://directory.google.com/Top/Computers/Artificial_Intelligence/Belief_Networks/Software/
The Bayesian Network Repository is also a valuable resource. It lists examples of Bayesian networks and datasets, from which they can be learned:
http://www.cs.huji.ac.il/labs/compbio/Repository/
The software we developed in connection with our book is available at:
http://fuzzy.cs.uni-magdeburg.de/books/gm/software.html
Tools for troubleshooting Microsoft products, which are based on Bayesian networks (but do not allow you to access them directly), can be found at
http://support.microsoft.com/support/tshoot/default.asp
back to the top
BayesBuilder
SNN, University of Nijmegen
PO Box 9101, 6500 HB Nijmegen, The Netherlands
http://www.mbfys.kun.nl/snn/Research/bayesbuilder/
BayesBuilder is a tool for (manually) constructing Bayesian networks and drawing inferences with them. It supports neither parameter nor structure learning of Bayesian networks. The graphical user interface of this program is written in Java and is easy to use. However, the program is available only for Windows, because the underlying inference engine is written in C++ and has only been compiled for Windows yet. BayesBuilder is free software.
back to the top
Bayesian Knowledge Discoverer / Bayesware Discoverer
Knowledge Media Institute / Department of Statistics
The Open University
Walton Hall, Milton Keynes MK7 6AA, United Kingdom
http://kmi.open.ac.uk/projects/bkd/
Bayesware Ltd.
http://bayesware.com/
The Bayesian Knowledge Discoverer is a software tool that can learn Bayesian networks from data (structure as well as parameters). The dataset to learn from may contain missing values, which are handled by an approach called "bound and collapse" that is based on probability intervals. The Bayesian Knowledge Discoverer is free software, but it has been succeeded by a commercial version, the Bayesware Discoverer. This program has a nice graphical user interface with some powerful visualization options. A 30 days trial version may be retrieved free of charge. Bayesware Discoverer is available for Windows, Unix and Macintosh.
back to the top
Bayes Net Toolbox
Kevin Patrick Murphy
Department of Computer Science, UC Berkeley
387 Soda Hall, Berkeley, CA 94720-1776, USA
http://www.cs.berkeley.edu/~murphyk/Bayes/bnt.html
The Bayes Net Toolbox is an extension for Matlab, a well-known and widely used mathematical software package. It supports several different algorithms for drawing inferences in Bayesian networks as well as several algorithms for learning the parameters and the structure of Bayesian networks from a dataset of sample cases. It does not have a graphical user interface of its own, but profits from the visualization capabilities of Matlab. The Bayes Net Toolbox is distributed under the Gnu Library General Public License and is available for all systems that can run Matlab, an installation of which is required.
back to the top
Belief Network Power Constructor
Jie Cheng
Dept. of Computing Science, University of Alberta
155 Athabasca Hall, Edmonton, Alberta, Canada T6G 2E1
http://www.cs.ualberta.ca/~jcheng/bnpc.htm
The Bayesian Network Power Constructor uses a three phase algorithm that is based on conditional independence tests to learn the structure of a Bayesian network from data. The conditional independence tests rely on mutual information, which is used to determine whether a (set of) node(s) can reduce or even block the information flow from one node to another. The program comes with a graphical user interface, though a much less advanced one than those of, for instance, HUGIN and Netica (see below). It does not support drawing inferences, but has the advantage that it is free software. It is available only for Windows.
back to the top
GeNIe / SMILE
Decision Systems Laboratory, University of Pittsburgh
B212 SLIS Building, 135 North Bellefield Avenue, Pittsburgh, PA 15260, USA
http://www2.sis.pitt.edu/~genie/
SMILE (Structural Modeling, Inference and Learning Engine) is a library of functions for building Bayesian networks and drawing inferences with them. It does support neither parameter nor structural learning of Bayesian networks. GeNIe (Graphical Network Interface) is a graphical user interface for SMILE, that makes the functions of SMILE easily accessible. While SMILE is platform independent, GeNIe is available only for Windows, since it relies heavily on the Microsoft Foundation classes. Both packages are distributed free of charge.
back to the top
Hugin
Hugin Expert A/S
Niels Jernes Vej 10, 9220 Aalborg, Denmark
http://www.hugin.com
Hugin is one of the oldest and best-known tools for Bayesian network construction and inference. It comes with an easy to use graphical user interface, but also has an API (application programmers interface) for several programming languages, so that the inference engine can be used in other programs. It supports estimating the parameters of a Bayesian network from a dataset of sample cases. In a recent version it has also been extended by a learning algorithm for the structure of a Bayesian network, which is based on conditional independence tests. Hugin is a commercial tool, but a demo version with restricted capabilities may be retrieved free of charge. Hugin is available for Windows and Solaris (Sun Unix).
back to the top
Netica
Norsys Software Corp.
2315 Dunbar Street, Vancouver, BC, Canada V6R 3N1
http://www.norsys.com
Like Hugin, Netica is a commercial tool with an advanced graphical user interface. It supports Bayesian network construction and inference and also comprises an API (application programmers interface) for C++, so that the inference engine may be used in other programs. Netica offers quantitative network learning (known structure, parameter estimation) from a dataset of sample cases, which may contain missing values. It does not support structural learning. A version of Netica with restricted capabilities may be retrieved free of charge, but the price of a full version is also moderate. Netica is available for Windows and Macintosh.
back to the top
Pulcinella
IRIDA, Université Libre de Bruxelles
50, Av. F. Roosevelt, CP 194/6, B-1050 Brussels, Belgium
http://iridia.ulb.ac.be/pulcinella/Welcome.html
Pulcinella is more general than the other programs listed on this page, as it is based on the framework of valuation systems [Shenoy 1992a]. Pulcinella supports reasoning by propagating uncertainty with local computations w.r.t. different uncertainty calculi, but does not support learning graphical models from a dataset of sample cases in any way. The current version of Pulcinella does not have a graphical user interface, but an outdated version of such an interface may be retrieved for Solaris (Sun Unix). Pulcinella is available for Solaris (Sun Unix) and Macintosh, but requires a Common Lisp system.
back to the top
Tetrad
Tetrad Project, Department of Philosophy
Carnegie Mellon University, Pittsburgh, PA, USA
http://hss.cmu.edu/html/departments/philosophy/TETRAD/tetrad.htm
Tetrad is based on the algorithms developed in [Spirtes et al 1993], i.e. on conditional independence test approaches to learn Bayesian networks from data, and, of course, subsequent research in this direction. It can learn the structure as well as the parameters of a Bayesian network from a dataset of sample cases, but does not support drawing inferences. Currently the program is being ported to Java (Tetrad IV). Older versions are available for MSDOS (Tetrad II) and Windows (Tetrad III). Tetrad II is commercial, but available at a moderate fee. Free beta versions are available of Tetrad III and Tetrad IV.
back to the top
WinMine / MSBN
Machine Learning and Statistics Group
Microsoft Research, One Microsoft Way, Redmond, WA 98052-6399, USA
http://research.microsoft.com/~dmax/WinMine/tooldoc.htm
WinMine is a toolkit, i.e. a set of programs for different tasks, rather than an integrated program. Most programs in this toolkit are command line driven, but there is a graphical user interface for the data converter and a network visualization program. WinMine learns the structure and the parameters of Bayesian networks from data and uses decision trees to represent the conditional distributions. It does not support drawing inferences. However, Microsoft Research also offers MSBN (Microsoft Bayesian Networks), a tool for (manually) building Bayesian networks and drawing inferences with them, MSBN comes with a graphical user interface. Both programs, WinMine as well as MSBN, are available for Windows only.
back to the top
© 2002 Christian Borgelt
Last modified: Fri Oct 25 11:05:52 MEST 2002
Subscribe to:
Posts (Atom)