What's New

Keep an eye on this page for major updates on what's been happening with annie.

Ondrej Pacovsky (OP): I have taken over annie because Asim is being slowly eaten by other projects. I have been working (a bit) on annie since Nov 2003 till now. My improvements are determined by the time available, too. I will implement only what I need. I'm trying to keep Annie coherent, robust and tested - but I have approx. 1/4 time needed to do that. So some parts are really just experimental and aren't easy and low-bugged to use. In the future, I will work on Recurrent nets, but most probably not on other TODOs.
0.60 and 0.70alpha releases are available. I would, however, suggest downloading a fresh CVS, because it will be probably a bit more stable at this stage. Features kohonen, examples with visualisation, interface tunes and... see the NEWS file
If you are a Windows user then please use version 0.51 instead of the new releases. Windows support for the new versions is not yet in place

There was a small error in the previous release which prevented it from compiling. If you're having compile trouble with release 0.51 then replace annie.h in that download with this file.
(Basically, there are three files that are NOT part of the release which were included in annie.h (these files were KohonenMap.h, KohonenMap1D.h and KohonenMap3D.h. To fix this error, just comment out the lines including these files in annie.h or download this file).

Minor bugfix - Compilation errors in Layer.h were observed when compiling in Visual Studio 6.0. This has been fixed
Visual Studio 6 - Workspace and solution files added.
This update ships as version 0.51. No changes to documentation or Matlab extensions pack, so they ship as in release 0.5. Furthermore, if you do not use Visual Studio 6, then you may not need to change to 0.51 from 0.5
PDF format documentation is now available for download

Matlab Extensions Pack - An add-on package has been developed to open the doors to the power of Matlab's Neural Network toolbox to annie! This pack will allow you to create training sets in C++, export them to matrices in Matlab, then create a network in Matlab, train it and then export it back so that you can use the network in your C++ applicaion! (README)
Hopfield Networks are now supported by annie. An example is provided that demonstrates the pattern storing ability of Hopfield networks.
Some bug fixes and other small new features. Fully compatible with older releases so you might as well just download it!

Last modified: Jun 18, 2004