librks ---- open source software for random kitchen sinks questions, bug report to olivier.teytaud@inria.fr 0) what is librks It's an implementation of random kitchen sinks. librks takes as input (i) a dataset (continuous domains) in the same format as libsvm/liblinear/gsll (ii) a set of kitchen sinks (a simple example is a set of gaussian random variables; typically one can use rg.txt provided with this source code) (iii) possibly, a number of sinks to be skipped. It outputs the same dataset, reformatted according to kitchen sinks. It is in particular suitable for learning with the following codes (same format at least in the non-sparse case) libsvm (http://www.csie.ntu.edu.tw/~cjlin/libsvm/ , i.e. SVM) liblinear (http://www.csie.ntu.edu.tw/~cjlin/liblinear/ , i.e. linear SVM for classification) gsll (http://www.lri.fr/~teytaud/mygsll , i.e. least squares linar regression) (see examples below) 1) download: you should find easily on www (google gsll gslp) (i) binary (statically linked) for linux 64 bits, linux 32 bits (ii) source code 2) compiling (if you downloaded the source) g++ -o librks librks.c 3) running You need a learning set (one is provided below for the example). Then, you can do as follows: ============================= example with gsll (http://www.lri.fr/~teytaud/mygsll) ============== ./librks 10 learningset rg.txt rkslearningset #### learning ./gsll rkslearningset model #### testing ./gslp rkslearningset model outputvalues ### here the learning set is equal to the ### test set - absolutely not required of course #### checking paste outputvalues learningset | awk ' { print $1 , " ", $2 }' #### prints the predicted and the real values ====================================================== 4) compatibility Should compile on any system Format compliant with libsvm / liblinear at least in the non-sparse case, but adapted for regression. 5) License This is under Cecill license (GPL-compliant). Learn more on http://www.cecill.info/. This comes with ABSOLUTELY NO WARRANTY, to the extent permitted by applicable law.