| Title: | Classification Based MCAR Test |
|---|---|
| Description: | Implementation of a KL-based (Kullback-Leibler) test for MCAR (Missing Completely At Random) in the context of missing data as introduced in Michel et al. (2021) <arXiv:2109.10150>. |
| Authors: | Meta-Lina Spohn [aut, cre], Loris Michel [aut], Jeffrey Naef [aut] |
| Maintainer: | Meta-Lina Spohn <[email protected]> |
| License: | GPL-3 |
| Version: | 1.0.1 |
| Built: | 2024-11-15 03:35:34 UTC |
| Source: | https://github.com/cran/PKLMtest |
Generate the test statistic
genU(st, lab)genU(st, lab)
st |
a ranger forest object. |
lab |
an integer value containing the class labels |
the likelihood-based test statistic
PKLMtest: compute a p-value for testing MCAR
PKLMtest( X, num.proj = 300, num.trees.per.proj = 10, nrep = 500, min.node.size = 10, size.resp.set = 2, compute.partial.pvals = FALSE, ... )PKLMtest( X, num.proj = 300, num.trees.per.proj = 10, nrep = 500, min.node.size = 10, size.resp.set = 2, compute.partial.pvals = FALSE, ... )
X |
a numeric matrix containing missing values encoded as NA, the data. |
num.proj |
a positive integer specifying the number of projections to consider for the score. |
num.trees.per.proj |
a positive integer, the number of trees per projection. |
nrep |
a positive integer, the number of permutations. |
min.node.size |
a positive number, the minimum number of nodes in a tree. |
size.resp.set |
an integer (>= 2), maximum number of classes allowed to be compared in each projection. |
compute.partial.pvals |
a boolean, indicate if partial p-values shopuld be computed as well. |
... |
additional parameters. |
a numeric value, the p-value(s) for the MCAR test, the first value is always the global p-value and if compute.partial.pvals is set to TRUE, the next values are the partial p-values for the relative importance of each variable.
n <- 100 X <- cbind(rnorm(n),rnorm(n)) X.NA <- X X.NA[,1] <- ifelse(stats::runif(n)<=0.2, NA, X[,1]) pval <- PKLMtest(X.NA, num.proj = 5)n <- 100 X <- cbind(rnorm(n),rnorm(n)) X.NA <- X X.NA[,1] <- ifelse(stats::runif(n)<=0.2, NA, X[,1]) pval <- PKLMtest(X.NA, num.proj = 5)
Truncation of probability
truncProb(p)truncProb(p)
p |
a numeric value between 0 and 1 to be truncated |
a numeric value with truncated probabilities