search for: 0.20000

I have the joint distribution of three discrete random variables z1, z2 and z3 which is captured by "z" and "prob" as described below. For example, the probability for z1=0.46667, z2=-1 and z3=-1 is 2.752e-13. Also, the probability adds up to 1. > head(z) z1 z2 z3 [1,] -0.46667 -1.0000 -1.0000 [2,] -0.33333 -0.9333 -0.9333 [3,] -0.20000 -0.8667 -0.8667

Most likely, previous computations have ended up giving slightly different values of say 0.13333. A pragmatic way out is to round to, say, 5 digits before applying unique. In this particular case, it seems like all numbers are multiples of 1/30, so another idea could be to multiply by 30, round, and divide by 30. -pd > On 28 Jul 2017, at 17:17 , li li <hannah.hlx at gmail.com> wrote:

Hi All, I have a data set wit three size classes (pico, nano and micro) and 12 different sites (Seamounts). I want to produce a table with the mean and standard deviation values for each site. Seamount Pico Nano Micro Total_Ch 1 Off_Mount 1 0.0691 0.24200 0.00100 0.31210 2 Off_Mount 1 0.0938 0.00521 0.02060 0.11961 3 Off_Mount 1 0.1130 0.20000 0.06620 0.37920 4 Off_Mount 1

Wolfgang, It is common to handle relative risk problems using Poisson regression. In your example you have 8 events out of 508 tries, and 0/500 in the second data set. > tdata <- data.frame(y=c(8,0), n=c(508,500), group=1:0) > fit <- glm(y ~ group + offset(log(n)), data=tdata, family=poisson) Because of the zero, the standard beta/se(beta) confidence intervals don't work.

I would like to create a tree model with at most one split point per variable using tree, rpart or other routine. Its OK if a variable enters at more than one node but if it does then all splits for that variable should be at the same point. The idea is that I want to be able to summarize the data as binary factors with the chosen split points. I don't want to have three level or more