kfold_ksmm: Parameter Tuning of Functions Using Grid Search

In kyuridata/ksmm: ksmm algorithm

Description

This generic function tunes hyperparameters of statistical methods using a grid search over supplied parameter ranges for ksmm.

Usage

	kfold_ksmm(x, y, x_dim, cost_range, sigma_range, kernel, nFold = 5, nCores = 1, ...) 

Arguments

x	a input data matrix
y	a response vector with one label for each row/component of X. It must be 1 or -1
x_dim	the size of input matrix X
cost_range	input data matrix to test
sigma_range	input data matrix to test
kernel	the kernel used in training and predicting
nFold	the number of partitions for cross-valication
nCores	the number of cores to use for parallel computing

Details

This function is used to tune the parameters of ksmm. Detailed theory is included in the KSMM paper.

Author(s)

Kyuri Park

References

Ye, Y. (2019). A nonlinear kernel support matrix machine. International Journal of Machine Learning and Cybernetics.

See Also

ksmm

Examples

require(ksmm)
data(nottingham)

X = as.matrix(nottingham[,-1])
y = ifelse(nottingham[,1] == 1, 1, -1)
cost_seq = 2^{-1:1}
sigma_seq = 10^{-1:1}
train_x = X[c(1,2,99,100),]
train_y = y[c(1,2,99,100)]

kfold_fit = kfold_ksmm(x= train_x, y = train_y, x_dim = c(200, 200), cost_range = cost_seq, 
sigma_range = sigma_seq, kernel='rbf', nCores = 1, maxit = 1e+3, epsilon = 5e-2, nFold = 1)
opt_cost = kfold_fit$opt_params[1]
opt_sigma = kfold_fit$opt_params[2]
    	

kyuridata/ksmm documentation built on Dec. 21, 2021, 8:47 a.m.