mixture.test: mixture.test
In jan-glx/knnDemix: Bound for mixing coefficient from two samples

Description Usage Arguments Details Examples

View source: R/knn_mixture_test.R

Tests against the true mixing coefficient of F_0 in the mixture F_m to be smaller than alpha using a knn statistic calculated from one sample from F0 and one sample from the mixture F_m.

1 2	mixture.test(X0, Xm, alpha = 1, alternative = "lower", conf.level = 0.95, k = 1, sub_sample = 1/4, seed_ = 1, calc.CI = TRUE)

`X0`	numeric matrix with rows beeing data points from a sample from F_0. Vector inputs are coerced to nx1 matrices.
`Xm`	numeric matrix with rows beeing data points from a sample from F_m. Vector inputs are coerced to nx1 matrices.
`alpha`	numeric value of the fraction of F_0 in F_m that is tested against
`alternative`	a character string specifying the alternative hypothesis - only 'lower' is allowed and only lower makes sense
`conf.level`	confidence level of the interval.
`k`	integer value indicating the which nearest neighbour is used for the testing density ratios.
`sub_sample`	numeric value specifing factor subsamplamping – only n*sub_sample_exp of all n pvalues will be used in order to reduce their depency
`seed_`	seed to be used for subsampling. Set to `NULL` if want reduce variance by averaging several subsamplings.
`calc.CI`	logical value indicating if confidence intervals for alpha schould be computed. Can be turend of to save computation time in case it is not needed.

F_m=α F_0 + (1-α) F_1 thus f_m=α f_0 + (1-α) f_1, f_1 >=0 thus f_m >= α f_0 ... TODO: finish explaination

n0 <- 100 # 100 sample points from \eqn{F_0}
nm <- 100 # 100 sample points from \eqn{F_0}
n1 <- 50  # of which 50 are from \eqn{F_1}
p <- 2    # two dimensions
X0 <- matrix(rnorm(n0*p), ncol=p) # \eqn{F_0} is standard normal, \eqn{F_1} has mean c(10,10):
Xm <- matrix(c(rnorm((nm-n1)*p), rnorm(n1*p)+10), ncol=p, byrow = TRUE)

mixture.test(X0, Xm, alpha=0.5, calc.CI=FALSE)$p.value #should not be significant
mixture.test(X0, Xm, alpha=1, calc.CI=FALSE)$p.value #should be significant
mixture.test(X0, Xm) # computes confidence intervals