Weighted sampling is the sampling process were each candidate has a corresponding weight, which determines its probability of being sampled. The weights are normalized, in order for their sum to equal one. Then, the normalized weights correspond to the probability that any individual will be sampled. For a simple example with three candidates, assuming weights of 1, 5, and 10, the following table depicts the normalized weights and the corresponding probability that any candidate will be chosen.
Candidate |
Weight |
Normalized weight |
Probability |
1 |
1 |
0.0625 |
6.25% |
2 |
5 |
0.3125 |
31.25% |
3 |
10 |
0.625 |
62.50% |
Instance weights to probabilities