Source codes of algorithms and datasets for our paper
"High-probability Kernel Alignment Regret Bounds for Online Kernel Selection",
accepted in ECML2021.

We implement all algorithms with R on a Windows machine with 2.5 GHz Core i7 CPU, 
execute each experiment 10 times with random permutation of all datasets and average all of the results.

The baseline algorithms include: BOGD, Forgetron, OKS and ISKA.
Our algorithms include: B(AO)2KS, EA2OKS and BEA2OKS.

The datasets are downloaded from: https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/
w7a (Num:24692, Fea: 300), 
w8a (Num: 49749, Fea: 300), 
a9a (Num: 48842, Fea: 123),
ijcnn1 (Num: 141691, Fea: 22).