cca via sparse smooth matrix prediction — smoothAppGradCCA • ANTsR

This implements a sparse and graph-regularized version of CCA based on the AppGrad style of implementation by Ma, Lu and Foster, 2015.

smoothAppGradCCA(
  x,
  y,
  smoox = NA,
  smooy = NA,
  sparsenessQuantile = 0.5,
  positivity = "either",
  k = 2,
  iterations = 10,
  stochastic = NA,
  initialization = "randxy",
  verbose = FALSE
)

Arguments

x: input view 1 matrix
y: input view 2 matrix
smoox: smoothingMatrix for x
smooy: smoothingMatrix for y
sparsenessQuantile: quantile to control sparseness - higher is sparser
positivity: restrict to positive or negative solution (beta) weights. choices are positive, negative or either as expressed as a string.
k: number of basis vectors to compute
iterations: number of gradient descent iterations
stochastic: size of subset to use for stocastic gradient descent
initialization: type of initialization, currently only supports a character randxy
verbose: boolean option

Value

list with matrices each of size p or q by k

Author

Avants BB

Examples


mat <- replicate(100, rnorm(20))
mat2 <- replicate(100, rnorm(20))
mat <- scale(mat)
mat2 <- scale(mat2)
wt <- 0.666
mat3 <- mat * wt + mat2 * (1 - wt)
jj <- smoothAppGradCCA(mat, mat3)