C-IL2P Changelog
What's new in C-IL2P 1.3 Beta
May 29, 2014- Several new functionalities have been added:
- A novel normalization that maintains C-IL2P building algorithm correctness has been made
- It is possible now to print network weight info after normalization and after training
- Now it is possible to run deduction datasets (no training data) and pure learning problems (no background knowledge data)
- More bugs and efficiency bottlenecks has been found and fixed
New in C-IL2P 1.2 Beta (May 29, 2014)
- Several new functionalities have been added:
- Pre-splitted data can now be used
- An optional stopping criteria has been added
- Input filtering with mRMR can be used to reduce input dimension
- Unknown features from training examples can now be complete with the initial network, after the building step
- More bugs and efficiency bottlenecks has been found and fixed
- The on-screen information is now more complete and intuitive
New in C-IL2P 1.1 Beta (Oct 6, 2012)
- Multi-class training/testing is now possible, by adding multiple heads into an example
- Several bugfixes and Linux / Windows compatibility issues have been solved