Graph clustering via LinLog force directed layout
Hi giraph community, I'm interested in using giraph for distributed n-body simulation. Initially, i'm interested in force directed layouts -- ie, graph drawing: http://en.wikipedia.org/wiki/Force-based_algorithms_(graph_drawing) I'm interested specifically in Dr. Andreas Noack's LinLog energy model -- which performs well w/ community detection: http://www.informatik.tu-cottbus.de/~an/GD/linlog.html I have a few examples of a serial implementation here: http://www.smarttypes.org/ The model maximizes the distance between all nodes while minimizing the distance between connected nodes. Without getting into too much detail, i'm curious if anyone has considered using giraph for force directed graph embedding (yet another name for it)? I'm also considering something like http://www.mcs.anl.gov/petsc/ or http://www.cs.cmu.edu/~scandal/alg/nbody.html -- which have fast n-body simulation implementations (Barnes-Hut + Fast Multipole). That said, i think giraph may be a good fit -- curious what the community thinks? Thanks, Timmy Wilson Cleveland, OH
Re: Graph clustering via LinLog force directed layout
Hi, I'm not definitely familiar with the algorithm or implementation of LinLog, I've been just a user. It should be doable with Giraph if you can express it in terms of message-passing between vertices and without a dependency on a global view of the graph (except for the convergence criteria, such as total energy). Please consider that Giraph's data model is based on a directed graph, this should be a quite interesting constraint for you, if your implementation is going to modify energy associated with edges (you'd have two views over the undirected edge, one in each endpoint). In general, a good way of doing community analysis would be to look at algorithms that belong to the family of label-propagation clustering algorithms. Hope this helps, Claudio On Tue, Mar 6, 2012 at 3:28 PM, Timmy Wilson tim...@smarttypes.org wrote: Hi giraph community, I'm interested in using giraph for distributed n-body simulation. Initially, i'm interested in force directed layouts -- ie, graph drawing: http://en.wikipedia.org/wiki/Force-based_algorithms_(graph_drawing) I'm interested specifically in Dr. Andreas Noack's LinLog energy model -- which performs well w/ community detection: http://www.informatik.tu-cottbus.de/~an/GD/linlog.html I have a few examples of a serial implementation here: http://www.smarttypes.org/ The model maximizes the distance between all nodes while minimizing the distance between connected nodes. Without getting into too much detail, i'm curious if anyone has considered using giraph for force directed graph embedding (yet another name for it)? I'm also considering something like http://www.mcs.anl.gov/petsc/ or http://www.cs.cmu.edu/~scandal/alg/nbody.html -- which have fast n-body simulation implementations (Barnes-Hut + Fast Multipole). That said, i think giraph may be a good fit -- curious what the community thinks? Thanks, Timmy Wilson Cleveland, OH -- Claudio Martella claudio.marte...@gmail.com
Re: Graph clustering via LinLog force directed layout
Hi Timmy, I don't know much about force directed layout, but it certainly sounds like a very interesting application for Giraph. Keep us posted on your progress and let us know how we can help. Avery On 3/6/12 8:34 AM, Claudio Martella wrote: Hi, I'm not definitely familiar with the algorithm or implementation of LinLog, I've been just a user. It should be doable with Giraph if you can express it in terms of message-passing between vertices and without a dependency on a global view of the graph (except for the convergence criteria, such as total energy). Please consider that Giraph's data model is based on a directed graph, this should be a quite interesting constraint for you, if your implementation is going to modify energy associated with edges (you'd have two views over the undirected edge, one in each endpoint). In general, a good way of doing community analysis would be to look at algorithms that belong to the family of label-propagation clustering algorithms. Hope this helps, Claudio On Tue, Mar 6, 2012 at 3:28 PM, Timmy Wilsontim...@smarttypes.org wrote: Hi giraph community, I'm interested in using giraph for distributed n-body simulation. Initially, i'm interested in force directed layouts -- ie, graph drawing: http://en.wikipedia.org/wiki/Force-based_algorithms_(graph_drawing) I'm interested specifically in Dr. Andreas Noack's LinLog energy model -- which performs well w/ community detection: http://www.informatik.tu-cottbus.de/~an/GD/linlog.html I have a few examples of a serial implementation here: http://www.smarttypes.org/ The model maximizes the distance between all nodes while minimizing the distance between connected nodes. Without getting into too much detail, i'm curious if anyone has considered using giraph for force directed graph embedding (yet another name for it)? I'm also considering something like http://www.mcs.anl.gov/petsc/ or http://www.cs.cmu.edu/~scandal/alg/nbody.html -- which have fast n-body simulation implementations (Barnes-Hut + Fast Multipole). That said, i think giraph may be a good fit -- curious what the community thinks? Thanks, Timmy Wilson Cleveland, OH
Re: Graph clustering via LinLog force directed layout
Hi Timmy, Sounds like a really cool idea to use giraph for layouting graphs, what is the complexity of that algorithm you plan to implement? --sebastian On 06.03.2012 22:29, Avery Ching wrote: Hi Timmy, I don't know much about force directed layout, but it certainly sounds like a very interesting application for Giraph. Keep us posted on your progress and let us know how we can help. Avery On 3/6/12 8:34 AM, Claudio Martella wrote: Hi, I'm not definitely familiar with the algorithm or implementation of LinLog, I've been just a user. It should be doable with Giraph if you can express it in terms of message-passing between vertices and without a dependency on a global view of the graph (except for the convergence criteria, such as total energy). Please consider that Giraph's data model is based on a directed graph, this should be a quite interesting constraint for you, if your implementation is going to modify energy associated with edges (you'd have two views over the undirected edge, one in each endpoint). In general, a good way of doing community analysis would be to look at algorithms that belong to the family of label-propagation clustering algorithms. Hope this helps, Claudio On Tue, Mar 6, 2012 at 3:28 PM, Timmy Wilsontim...@smarttypes.org wrote: Hi giraph community, I'm interested in using giraph for distributed n-body simulation. Initially, i'm interested in force directed layouts -- ie, graph drawing: http://en.wikipedia.org/wiki/Force-based_algorithms_(graph_drawing) I'm interested specifically in Dr. Andreas Noack's LinLog energy model -- which performs well w/ community detection: http://www.informatik.tu-cottbus.de/~an/GD/linlog.html I have a few examples of a serial implementation here: http://www.smarttypes.org/ The model maximizes the distance between all nodes while minimizing the distance between connected nodes. Without getting into too much detail, i'm curious if anyone has considered using giraph for force directed graph embedding (yet another name for it)? I'm also considering something like http://www.mcs.anl.gov/petsc/ or http://www.cs.cmu.edu/~scandal/alg/nbody.html -- which have fast n-body simulation implementations (Barnes-Hut + Fast Multipole). That said, i think giraph may be a good fit -- curious what the community thinks? Thanks, Timmy Wilson Cleveland, OH
Re: Graph clustering via LinLog force directed layout
Thank you everyone! I would love to see a comparison of force directed layouts (specifically LinLog) and label propagation. I searched but alas nothing -- they seem to be oddly similar? The current, serial LinLog implementation -- http://code.google.com/p/linloglayout/ -- uses Barnes-Hut simulation -- n*log(n): http://en.wikipedia.org/wiki/Barnes%E2%80%93Hut_simulation I guess the root question is -- do you think it's reasonable to use giraph for Barnes-Hut simulation? On Tue, Mar 6, 2012 at 11:34 AM, Claudio Martella claudio.marte...@gmail.com wrote: Hi, I'm not definitely familiar with the algorithm or implementation of LinLog, I've been just a user. It should be doable with Giraph if you can express it in terms of message-passing between vertices and without a dependency on a global view of the graph (except for the convergence criteria, such as total energy). Please consider that Giraph's data model is based on a directed graph, this should be a quite interesting constraint for you, if your implementation is going to modify energy associated with edges (you'd have two views over the undirected edge, one in each endpoint). In general, a good way of doing community analysis would be to look at algorithms that belong to the family of label-propagation clustering algorithms. Hope this helps, Claudio On Tue, Mar 6, 2012 at 3:28 PM, Timmy Wilson tim...@smarttypes.org wrote: Hi giraph community, I'm interested in using giraph for distributed n-body simulation. Initially, i'm interested in force directed layouts -- ie, graph drawing: http://en.wikipedia.org/wiki/Force-based_algorithms_(graph_drawing) I'm interested specifically in Dr. Andreas Noack's LinLog energy model -- which performs well w/ community detection: http://www.informatik.tu-cottbus.de/~an/GD/linlog.html I have a few examples of a serial implementation here: http://www.smarttypes.org/ The model maximizes the distance between all nodes while minimizing the distance between connected nodes. Without getting into too much detail, i'm curious if anyone has considered using giraph for force directed graph embedding (yet another name for it)? I'm also considering something like http://www.mcs.anl.gov/petsc/ or http://www.cs.cmu.edu/~scandal/alg/nbody.html -- which have fast n-body simulation implementations (Barnes-Hut + Fast Multipole). That said, i think giraph may be a good fit -- curious what the community thinks? Thanks, Timmy Wilson Cleveland, OH -- Claudio Martella claudio.marte...@gmail.com
