All notable changes to LIBTwinSVM library will be documented in this file. The format is based on Keep a Changelog and this project adheres to Semantic Versioning.
- Checking whether the kernel names are valid.
- The order of computations is optimized in TSVM and LSTSVM estimators to reduce their memory consumption.
- The
fit()method of the TSVM-based estimators now accepts a Python list as well as NumPy arrays.
- The optimizer component avoids memory copies. It substantially improved the training speed and memory usage of the TSVM estimator.
- Fixed the heap corruption of the optimizer component for Windows platforms.
- Added two benchmark scripts to the repository.
- Added Grid Search method to the Library's API.
- Added two more usage examples to the documentation of the project.
- Fixed adding LAPACK and BLAS DLLs for Windows platform.
- Fixed an Installation error for Linux systems.
- A workaround for installing the library on MacOSX systems.
- To overcome matrix singularity of linear LSTSVM, an extra stabilizer term added to the equations.
- A graphical user interface (GUI) application.
- Fast implementation of standard TwinSVM and Least Sqaures TwinSVM.
- A fast optimizer (ClipDCD) in Cython.
- Implemented multi-class schemes (One-vs-All & One-vs-One) as meta-estimators.
- A feature-rich visualization tool to show decision boundaries.
- A module for saving and loading TSVM-based models.
- A setup.py file for installing LIBTwinSVM as a Python package.