# NOT RUN {
data(benchmark.data)
data.mkl=benchmark.data[[1]]
kernels=rep('radial',2)
sigma=c(2,1/20)
train.samples=sample(1:nrow(data.mkl),floor(0.7*nrow(data.mkl)),replace=FALSE)
degree=sapply(1:length(kernels), function(a) ifelse(kernels[a]=='p',2,0))
#Kernels.gen splts the data into a training and test set, and generates the desired kernel matrices.
#Here we generate two gaussisan kernel matrices with sigma hyperparameter 2 and 0.05
K=kernels.gen(data=data.mkl[,1:2],train.samples=train.samples,
kernels=kernels,sigma=sigma,degree=degree,scale=rep(0,length(kernels)))
C=0.05 #Cost parameter for DALMKL
K.train=K$K.train
K.test=K$K.test
ytr=data.mkl[train.samples,3]
#Converts list of kernel matrices in to an array with is appropriate for C++ code
k.train=simplify2array(K.train)
k.test=simplify2array(K.test)
#Implement DALMKL with the hinge loss function
spicy_svmb1n=SpicyMKL(K=k.train,y=ytr, loss='hinge',C=C)
prediction_logistic=predict_Spicy(spicy_svmb1n$alpha,spicy_svmb1n$b,k.test)
#Implement DALMKL with the hinge loss function
spicy_logistic=SpicyMKL(K=k.train,y=ytr, loss='logistic',C=C)
prediction_logistic=predict_Spicy(spicy_logistic$alpha,spicy_logistic$b,k.test)
# }
Run the code above in your browser using DataLab