statnet
a feedforward interpolation neural network

statnet is a feedforward neural network (multi-layer perceptron) for implementing non-linear static functions. It is trained by backpropagation. The network incorporates the following features:

• committees of networks
• weight decay
• noise levels (precision setting) for each target variable
• feature selection (i.e. selection of subsets of input and output variables from the data files)
• pattern selection
• mean and variance scaling of input and output variables
• direct or logged target variables
• conjugate gradient or gradient descent optimization
• ability to have undefined (missing) targets in training data file
• one or two hidden layers
• tanh input-hidden transfer function, linear hidden-output function
• sum-of-squares error measure

statnet can be downloaded here. You are free to use this code for academic and educational purposes provided that (1) you acknowledge its use and authorship, and (2) you inform me that you are using it. This way I can also let you know of any changes and improvements. The code is written in ANSI C. The file is 1Mb gzipped TAR file (it's large because of the test data supplied with it). So to extract do:
gunzip statnet.tar.gz
tar xvf statnet.tar
This will place all of the files in a directory called "statnet". Please read the obligatory README file. The file "statnet_manual" contains all you need to know to get started. Get in touch if you have any problems.

Download current version: (v2.00, 24.06.99)

Online manual