It is equivalent to compute K using the normalization `X/sqrt(sum(X^2))` in Feature Space.
References
Ah-Pine, J. (2010). Normalized kernels as similarity indices. In Advances in Knowledge Discovery and Data Mining: 14th Pacific-Asia Conference, PAKDD 2010, Hyderabad, India, June 21-24, 2010. Proceedings. Part II 14 (pp. 362-373). Springer Berlin Heidelberg. Link
Examples
dat <- matrix(rnorm(250),ncol=50,nrow=5)
K <- Linear(dat)
cosNorm(K)
#> [,1] [,2] [,3] [,4] [,5]
#> [1,] 1.00000000 -0.18576970 0.14211804 0.04982009 0.17898714
#> [2,] -0.18576970 1.00000000 0.09582528 -0.20726749 -0.04930542
#> [3,] 0.14211804 0.09582528 1.00000000 -0.17213295 -0.05003129
#> [4,] 0.04982009 -0.20726749 -0.17213295 1.00000000 0.21781431
#> [5,] 0.17898714 -0.04930542 -0.05003129 0.21781431 1.00000000