Not my area of expertise, but based on what I've read, you general try to get as close as possible through pure mappings, and then use rejection sampling to thr

Not my area of expertise, but based on what I've read, you general try to get as close as possible through pure mappings, and then use rejection sampling to throw out the samples that don't fall under the desired distribution. This can work even if the distribution is dynamically changing.