Legacy versions
The previous toolbox generations are preserved here for citation targets and reproducibility of published results. Both are unmaintained and unsupported; new work should use FST4. The same terms apply as for FST4: non-commercial use, no warranty, no liability.
Reproducibility note. FST4 ships parity configs reproducing every published FST3 demo protocol — but the FST3-era speech data set left the FST4 distribution and the sonar data's class split was corrected, so FST3-published numbers for those demos are reproducible only from the original FST3 package below with its original data.
FST3 (2010–2012)
The C++ library generation: templated C++98 code built on Boost, publishing the complete search family (sequential, floating, oscillating, dynamic oscillating, retreating; exhaustive and four Branch & Bound variants; Monte Carlo; BIF and DAF ranking) with filter and wrapper criteria, nested data splitting, result regularization, and stability evaluation. Final version: 3.1.1.beta (September 2012); see History for the version record. The package contains its own documentation and 36 example programs.
Reference: Petr Somol, Pavel Vácha, Stanislav Mikeš, Jan Hora, Pavel Pudil, and Pavel Žid. Introduction to Feature Selection Toolbox 3 — The C++ Library for Subset Search, Data Modeling and Classification. UTIA Tech. Report No. 2287, 2010 (PDF). (A poster summarizing FST 3.1 is also available: PDF.)
By downloading it you acknowledge that FST3 is provided “as is”, without warranty of any kind and without support, that the authors accept no liability for any damages arising from its use, and that it is licensed for non-commercial use only.
FST1 (1999–2002)
The first installment: a Windows executable with a graphical user interface, suitable primarily as an educational tool for experimenting with feature selection methods. Some of its methods (feature selection based on normal mixture modeling with the EM algorithm, TRN data file manipulation, limited visualization) were never re-implemented in the later generations. Compared to FST3/FST4 it supports no wrappers, no configurable data splitting, and no threading; it has not been developed since 2002 and may not run on current Windows versions.
Documentation: FST1 manual (PDF).
References: Petr Somol and Pavel Pudil. Feature Selection Toolbox. Pattern Recognition, 35(12):2749–2759, 2002 · Petr Somol and Pavel Pudil. Feature Selection Toolbox as a Multi-Purpose Tool in Pattern Recognition. In PRIS 2001, pages 91–102, ICEIS Press, 2001.
Download FST1 (Windows binary, zip)
By downloading it you acknowledge that FST1 is provided “as is”, without warranty of any kind and without support, that the authors accept no liability for any damages arising from its use, and that it is licensed for non-commercial use only.