search for: stuv

...rpolation as a first cut. 2.) find the subset of points that have more than one neighbour - this is kernel density estimation with a boxcar window of half-width del 3.) perform multinomial sampling on the x values for that subset - ie from (abc)(defg)(hij) select, say, afj, with probabilities (pqr)(stuv)(wxy) Step 2 could be done with a call to outer(), but that would probably be wasteful, especially 'cos it doesn't exploit the ordering of the values. A C program that runs through the sorted values would be quick. As for 3, sample() does multinomial sampling for a single set of p values...