ANNA 1.8 28-04-2008
what's ANNA?
ANNA: (Artificial Neural Network Architecture) is a Back propagation neural network class developed to fit the FLTK library v1.1.8. The training algorithms include the Conjugate Gradient and the Levenberg - Marquardt. The distribution include the source code and a demo for Linux systems. The structure is very flexible and you can change the number of inputs, number of layers, number of neurons per layer and the outputs in a simple way. There is included a very nice structure editor, so you can visualize the neuronal network structure. It is distributed under the GNU GPL's, or FLTK licenses.
The input data is a very simply matrix in a plain text file format, so you just write your data in the same way. The output are taken from the last column toward the first. So if the matrix has four columns you can take as much the last three ones as outputs, leaving the others as inputs.
ANNA screen shots (old version)
Functionalities of the graphic interface:
+ Structure editor
+ Training graphs
+ The training algorithms: Conjugate Gradient and the
Levenberg-Marquardt
+ Multiple outputs
+ Small tutorial of how carry out the Jacobian matrix calculation, it is still in early stage, all comments and
corrections are welcome jabcobian_ann.pdf.
Some fixed problems:
+ Problems with Table and Cartesian were fixed
+ Fixed an error of the scaling output function
+ More robust lecture of data file.
To compile in a 64bits machine try changing the line in the makeinclude file
LX11R = -L /usr/X11R6/lib -lGLU -lGL -lXext -lX11
by:
LX11R = -L /usr/X11R6/lib64 -lGLU -lGL -lXext -lX11
The input data is a very simply matrix in a plain text format file, so you just need to write your own data in the same way, the outputs are counted from the last column to the first, so if the matrix has four columns you can take as much the last three ones as outputs, because you have to leave some for input.
Ok, I hope it will help you,
Thanks, Edmanuel
Please, send me any question or feature to eetorres@gmail.com