When `length( skewed.probs ) > 200' uniform samples are generated in R-devel.
R-3.5.1 behaves as expected. `epsilon` can be a lot bigger than illustrated and still the uniform distribution is produced. Chuck > set.seed(123) > > epsilon <- 1e-10 > > ## uniform to 200 then small > p200 <- prop.table( rep( c(1, epsilon), c(200, 999-200))) > ## uniform to 201 then small > p201 <- prop.table( rep( c(1, epsilon), c(201, 999-201))) > > brks <- c(0,99,199,200,201,Inf) > tab200 <- sample( length(p200), 10000, prob=p200, replace=TRUE) > tab201 <- sample( length(p201), 10000, prob=p201, replace=TRUE) > > cbind( + s200=table(cut(tab200, brks)), + p200=round(xtabs(p200 ~ cut( seq_along(p200), brks)) * 10000 ,1), + s201=table(cut(tab201, brks )), + p201=round(xtabs(p201 ~ cut( seq_along(p201), brks)) * 10000 ,1)) s200 p200 s201 p201 (0,99] 5017 4950 984 4925.4 (99,199] 4925 5000 959 4975.1 (199,200] 58 50 9 49.8 (200,201] 0 0 6 49.8 (201,Inf] 0 0 8042 0.0 > > > > > sessionInfo() R Under development (unstable) (2019-03-02 r76189) Platform: x86_64-apple-darwin18.2.0 (64-bit) Running under: macOS Mojave 10.14.3 Matrix products: default BLAS: /Users/cberry/projects/R/R-devel/lib/libRblas.dylib LAPACK: /Users/cberry/projects/R/R-devel/lib/libRlapack.dylib locale: [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8 attached base packages: [1] stats graphics grDevices utils datasets methods base loaded via a namespace (and not attached): [1] compiler_3.6.0 > ______________________________________________ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel