Stable and production tested FFNN package

With this release we reached some maturity, providing a relatively well-tested FFNN library that suits our main use case well, is quite fast for small-medium sized networks and already provides some advanced functionality like a basic set of feature maps. Nevertheless, a lot more improvements are possible and will be implemented in the near future.

Code used for NNVMC article (2019)

This is a snapshot of the FFNN library as it was when employed for production of data for the (to be published) article about NNVMC by J. Kessler, F. Calcavecchia and T.D. Kühne in early 2019.

Although it was used successfully in production, this snapshot is not necessarily stabilized and well documented for public use. A better polished version is intended for the future and will be "released" as v0.2 .

Basic FFNN + Trainer

v0.1

• clean and stable basic FFNN with well-separated component class structure
• includes most common activation functions
• provides derivatives d/dx, d²/dx², d/dv, d²/dv/dx, d³/dv/dx²
• is reasonably performance optimized
• handles least-squares-fitting (training) against target data with minimal hassle for the user
• allows data split in training/validation/testing and early stopping technique