- Corrected default value of rho in sigmoid function from 0.5 to (the more conventional) 1.0.
- Fixed minor bugs in the training scripts.
- Re-trained all the examples to use the sigmoid function's new default settings.
- Added "Tanks a lot" example to the documentation.
- Removed the unused
tlufunction. - Updated tests to reflect this change.
- Minor documentation tidy-ups.
- API naming clarification:
- Renamed
clean_anntoclean_network. - Renamed
create_anntocreate_network.
- Renamed
- Documentation refinement.
- Initial release.
- Docs at https://sann.readthedocs.org/.
- Full test coverage.
- Two examples:
- digit recognition - character recognition via supervised backpropagation training.
- snAIke - play the SNAKE game via unsupervised neuro-evolution training.
- Blog post exploring the concepts: Behind the AI Curtain