tests/test-QYgX.R

In vincenzocoia/copsupp: Vine Copulas made easy

# ## Check that QYgX() gives the same result as rvineqcond()
# library(CopulaModel)
# library(copsupp)
# library(cnqr)
# library(testthat)
# data(egdat)
# xyinfo <- xylink(egvine)
#
# ## Useful quantities
# A <- GtoA(egvine$G)
# parvec <- c(t(egvine$cparmat), recursive=TRUE)
# np <- apply(egvine$cparmat, 1:2, function(l) length(l[[1]]))
# qcondmat <- apply(egvine$copmat, 1:2, function(cop) {
#     if (cop != "") return(paste0("qcond", cop)) else return("")
# })
# pcondmat <- apply(egvine$copmat, 1:2, function(cop) {
#     if (cop != "") return(paste0("pcond", cop)) else return("")
# })
#
# ## Check pcondseq
# uind_pcondseq <- pcondseq(egdat, ord=1:5, rv=egvine)
# uind_rvinepcond <- rvinepcond(parvec, egdat, A=A, ntrunc=4, pcondmat=pcondmat, np=np)
# expect_identical(uind_pcondseq, uind_rvinepcond)
#
# ## Compute quantiles
# ## NOTE: rvineqcond() computes the (5|1:4)-quantiles. That is, the quantile
# ##  levels used for the i'th observation is the i'th PIT score of 5|1:4.
# ## NOTE2: We need 'uind' here, separate from uind_pcondseq, because the
# ##  former has PIT scores (4), (3|4), (2|3:4), (1|2:4), which are needed in the
# ##  computation of the 5|1:4 quantiles. uind_pcondseq has PIT scores
# ##  (1), (2|1), (3|1:2), (4|1:3), which are not needed in the computation of
# ##  the 5|1:4 quantile.
# uind <- pcondseq(egdat, ord=xyinfo$xord, rv=egvine)
# yhat_QYgX <- sapply(1:length(vcond), function(i)
#     QYgX(uind_pcondseq[i, 5], uind, cops=xyinfo$cops, cpars=xyinfo$cpars, QY=identity)[i, ])
# yhat_rvineqcond <- rvineqcond(uind_pcondseq, A, ntrunc=4,
#                               parvec=parvec,
#                               np=np,
#                               qcondmat=qcondmat,
#                               pcondmat=pcondmat)[, 5]
# expect_equal(yhat_QYgX, yhat_rvineqcond)  # They're equal, but not identical.
# max(abs(yhat_QYgX - yhat_rvineqcond))