Feature request: Add a Poisson sampling option. In other words, add a weight option to sampling to allow unequal probabilities of each row being sampled.

[BorgeJorge]

Poisson sampling is a generalization of Bernoulli sampling in which each row can have a different probability of being sampled. If the sampling procedure could accept a numeric variable as a sampling weight, the probability of any given row being sampled could then be calculated as the sampling weight for that row divided by the total sum of weights (with zero-weighted rows having zero probability of being sampled). This seems like a straightforward extension of Bernoilli sampling, in which the probability of a row being sampled is 1/total N of rows, which is equal to the Poisson sampling formula when all rows have a sampling weight of 1.

This would enable us to pull samples with specific distributions of characteristics, which would greatly benefit some kinds of exploratory analysis.