This library comes as a .msi (Microsoft installer) and a license in .rtf format ("rich text" -- who uses this anymore?). In short, it's what you'd expect from Microsoft: not open source, not open at all, and the license probably limits you to non-commercial use and non-redistribution.
Aforge is not even close to Infer.NET. The latter is a library to build graphical models and automatically generate deterministic inference algorithms for them. Trust me when I say there are probably less then 25 people in the world that can derive some of these inference algorithms without spending weeks studying the relevant papers (e.g. EP).
Aforge is just a collections of algorithms. It's nothing to do with building your own models. Useful nonetheless.
MS Cambridge has been developing this for years. I have only done some toy tests on it a couple of years ago. I think it could be very useful, but that license...ouch! So unless you are an academic, student, hobbyist, or (maybe) charity, why bother?
It comes with source, and is open-source for non-commercial purposes. (It's not licensed for commercial use at all.)
I'm not convinced this is so bad. Most companies who don't want to share in-house software with other companies just keep it in-house, and the world at large never hears of it. By sharing the software with academics and hobbyists, they're at least sharing it more than the default not-at-all.
It's not open source according to "The Open Source Definition" http://www.opensource.org/osd.html - being able to simply view the source code is just not good enough.
None of your comment has anything to do with the rest of your comment.
Seriously, how do you get all that from "uses platform-standard installer and file formats"? Besides which your assumptions seem to be answered right on the download page....
Includes the Infer.NET compiler and runtime, examples and tutorials, offline documentation, and source code for factors and distributions.
Please note that Infer.NET is downloadable for non-commercial use only; no commercial licenses are available at this time.
The tutorial is an nice rapid summary on how to use this to drive some models and understand modern bayesian inference. Worth taking a look at, if you don't want to read the ".rtf" info...
Are there any other recommendable ML libraries? I know of Mahout and have read up on it, and have been reading "Programming Collective Intelligence", but am eager to learn more.
I used this one summer to implement a graphical model I was working with. It has two approximate inference algorithms: variational message passing and expectation propagation, for which the inventors are co-authors of this library. I found the library to be powerful for those who knew what they were doing. But it took me a while to figure out the syntax.
Some bits were a bit hack-ey which made the syntax difficult for me to get right. For example, random variables are templates that take a bool or float depending on the underlying distribution. You could have a C# array of random variables, or a infer.net array of random variables, which had slightly different functionality. The infer.net array of random variables did not use the C# for loop, and instead had its own for loop implemented as a method.
Anyways, it's a nice library and great that they released the source code.
16 comments
[ 3.0 ms ] story [ 46.7 ms ] threadhttp://code.google.com/p/aforge/
Aforge is just a collections of algorithms. It's nothing to do with building your own models. Useful nonetheless.
I'm not convinced this is so bad. Most companies who don't want to share in-house software with other companies just keep it in-house, and the world at large never hears of it. By sharing the software with academics and hobbyists, they're at least sharing it more than the default not-at-all.
Seriously, how do you get all that from "uses platform-standard installer and file formats"? Besides which your assumptions seem to be answered right on the download page....
Includes the Infer.NET compiler and runtime, examples and tutorials, offline documentation, and source code for factors and distributions.
Please note that Infer.NET is downloadable for non-commercial use only; no commercial licenses are available at this time.
http://research.microsoft.com/en-us/um/cambridge/projects/in...
Are there any other recommendable ML libraries? I know of Mahout and have read up on it, and have been reading "Programming Collective Intelligence", but am eager to learn more.
Some bits were a bit hack-ey which made the syntax difficult for me to get right. For example, random variables are templates that take a bool or float depending on the underlying distribution. You could have a C# array of random variables, or a infer.net array of random variables, which had slightly different functionality. The infer.net array of random variables did not use the C# for loop, and instead had its own for loop implemented as a method.
Anyways, it's a nice library and great that they released the source code.