The new version (0.998) comes with a java GUI (lwnn_jtrainer) which should be much easier to use than lwnn_visualtrainer in TCL/TK. A new training utility, lwnn_xmltrain, uses a XML configuration file to set the training parameters.
Next step will be the porting to Windows.
This new version is a bugfix version of a problem that appeared with new versions of bash in the script that calls gnuplot in order to plot the error
This is just a bugfix version
Download if you've had problem compiling the previous version
In the homepage of the project you can find the english version of the report. It can be useful for programmers, for simple users, or for students who want to study neural networks
direct link:
http://lwneuralnetplus.sourceforge.net/report/lwnnplus_report_en.ps.gz
This versione makes some steps in the direction of portability. Will the library ever be compiled for M$ Windows?
Klwnn is an interface for KDE for our lwnnplus library. This is the first version of Klwnn. It depends on lwnnplus library and qwt library, that can be found at
http://sourceforge.net/projects/qwt/
This new version should be ANSI C++ (maybe some problems with namespaces?).
A new utility to convert digits in ocr example in pgm images
In version 0.992 iomanagebinary class allows you to manage patterns in binary form and to convert from every format to binary
With this release the project seems to be stable.
trainer class is now easier to use.
We are working on a qt application which uses the library
Trainer class, graphical interface
Possibility to read patterns in format of fann library
http://fann.sourceforge.net/
In this version we added a new class "trainer" that can be used to easily train a network in many ways.
There is also a graphical interface in tcl/tk for network training
In this new release some bugs have been fixed. There are no new features in the library itself but the program which generates interpolation of sigmoidal functions (generate_sigma) has now the possibility to switch the values for tanh function
Maybe we found the right way to use SuperSab (in combination with batch training).
Seems to be very fast
See train_network_ssab_batch in the examples in directory tests/
In this version some bugs have been corrected in batch training and in examples in directory tests
Documentation in PDF has been released in the package lwneuralnet++_doc
The new version has just some errors in documentation corrected.
You can find the documentation visiting the home page of the project.
Documentation:
http://lwneuralnetplus.sourceforge.net/docs/classnetwork.html
The first version of lwneuralnet++ is out.
The interface of the class is much like C
lwneuralnet++ supports momentum in batch training and SuperSab mode.
It needs testing!