Warning

This software is not usable for now, I just pushed the raw source code of a quick tool that I made for my experiments. I plane to reengineer it soon.

General informations

This program is a command-line tool that provides an easy way of learning, testing and running artificial neural networks (ANN). It uses the Fast Artificial Neural Network library (FANN) for learning and testing.

What are artificial neural networks ?

An artificial neural network (ANN), usually called "neural network" (NN), is a mathematical model or computational model that tries to simulate the structure and/or functional aspects of biological neural networks. An ANN performs a non-parametric non-linear multivariate multiple regression.

What is Fast Artificial Neural Network library ?

The Fast Artificial Neural Network library (FANN) is a free open source neural network library, which implements multilayer artificial neural networks in C and supports both fully and sparsely connected networks. Cross-platform execution in both fixed and floating point is supported. It includes a framework for easy handling of training data sets. It is easy to use, versatile, well documented, and fast. PHP, C++, .NET, Ada, Python, Delphi, Octave, Ruby, Prolog Pure Data and Mathematica bindings are available.

News

• 2010.05.08: support of the Continuous and Labels Icsiboost data fields (option -S)
• 2010.04.19: initial release

Features of the program

The main features of this program are :

• Support of Icsiboost / Boostexter file format (see ProgramUsage)
• Support of the native Fann data format
• Designed for resolving classification problems, error measurement is done in terms of classification error rate
• Automatic cross-validation evaluation
• Auto-saving of ANN that performs the best on train, dev or test

Please take a look at the ProgramUsage page for more details.

TODO - Not yet implemented (but should be soon)

• support of the Text Icsiboost data field
• selection of the training algorithm (currently it uses RPROP)
• selection of the number of layers (currently it produces 3 layer MLPs)
• selection of activation functions (currently it uses symmetric sygmoids for both hidden and output layers)
• selection of network density (currently it produces full connected MLPs)
• selection of error mode (currently only classification mode)

Getting and compiling

svn checkout http://sfann.googlecode.com/svn/trunk/ sfann
cd sfann
./configure
make