The theory behind sampling from Zipfian distributions, and why the inverse CDF is not always enough.
Understanding Rejection–Inversion Sampling for Zipf Distributions
The theory behind sampling from Zipfian distributions, and why the inverse CDF is not always enough.
The inverse CDF gives a direct way to sample from a distribution, but for discrete monotone distributions it can be expensive to evaluate. Rejection inversion trades a little rejection for a much cheaper proposal step.
You accept a proposal with a probability that corrects the distribution. The result is exact, provided the acceptance test is implemented carefully.
The best sampler is the one that is easy to reason about under load.
Cache Eviction Policies in Storage Engines
Caching stores frequently accessed items in fast volatile storage. The interesting part is deciding what to evict, and when.
Read articleGenerating Zipfian Distributions Using Rejection Inversion
Rejection inversion is a sampling method for discrete monotone distributions. Here is how to use it to benchmark and test software.
Read articleCumulative Distribution Function and Inverse CDF
What CDF and inverse CDF mean, and how they turn a uniform random number into a sample from any distribution.
Read article