Dear Gneural Network community,
I have worked quite some time on the library now, and I have researched
the feasibility of a new algorithm to train the network and calculate
output statistics based on bayesian statistics. It works well with both
curve fitting tests, although it is still quite slow. However, it can be
easily parallelized, so it is to be expected that improvements are still
possible.
Unfortunately, due to the intransparent development and the huge change
in the codebase (I had to write a whole new framework for this feature),
I have developed a fork of version 0.8 and I have not ported it back to
gneural network yet. However, I would like to share my findings with
you. For that, I have set up a GitHub repository , which you can access
at https://github.com/tobwes/bayesiannetwork and I would be happy if you
give your comments and ideas regarding this feature. Please also read
the README.md file in that git repository.
If you are interested, I am willing to port this back into the gneural
network codebase, but first I'll invite you to play with this piece of
software.