JavaNNS - Parsing the created Neural Network

I'm currently working on a Neural Network for creating a "better" PNG Predictor (Prefilter). I already create开发者_开发知识库d the network (with JavaNNS) which has a quite good learning rate on 8-Bit grayscale images.

Now my next step would be to include this created network in my prepared PNG Encoder/Decoder which is written in Java. But to do that I need to parse the created .net file from JavaNNS.

I don't want to invent the wheel again so is there any chance that another one of you has already written a simple parser for the .net files of the JavaNNS which would read all the layers with the neurons, the connections and the weights on the connections and store it in any usable Java data structure?

I know it isn't that hard to create a parser, but it would be awesome to save time and skip this "boring" task.. :)

Thanks!

JavaNNS's predecessor, SNNS (alternative link), had an export function which exported the trained network as C code (essentially a header and source file pair). These files could then be used in custom code.

In Java you could use JNI or JNA to call C code and I am sure there are threads here on SO how to accomplish this.

I do not know the current version of JavaNNS, but maybe they already provide an export function exporting the network as Java code instead of C? Or you could open your trained network in the old SNNS and export it?

I just wanna add that I've created my own .net (JavaNNS) File Parser in Java. It is possible to parse single hidden layer networks. If anyone needs the code who run into the same problem I had before I am happy to share my code.

You can contact me on my blog. Here is the post about the Neural Network Project I've done. Just let a comment there and I will provide you the JavaNNS Parser and the corresponding NeuralNetwork Class.

http://prineblog.wordpress.com/2011/06/21/neural-network-as-predictor-for-image-coding-png/

I just updated the Version of Nen to Beta - It is a lightweight 3-layer neural network implementation in Java for regression and classification. A little performance comparison against support vector machines (LibSVM) demonstrates its capabilities. It can be used via command line or Java.