C Deeply (cdeeply.com) builds neural networks very quickly from small training datasets (<100 kB). The web portal and APIs allow a 1 second computation time, used for architecture search and network training.
This repository contains the APIs that allow C Deeply's algorithms to be called directly by another application rather than manually through web form submissions. The current interfaces are:
- C / C++
- MATLAB / Octave
- Python
- R
- UNIX Bash scripts -- these embed neural networks in myNN.html, just as the web portal does
Each interface gives access to two methods:
- Supervised training (regression), used to predict certain columns/rows from others in a numeric table.
- Unsupervised training to find features in tabular data. The algorithm can produce an autoencoder, or just an encoder or decoder.
Additional methods (e.g. reinforcement training, methods to use on image/audio data, etc) will be added as they are developed.
Questions or comments should be directed to: info@cdeeply.com.