Graduate Descent

Gumbel-max trick and weighted reservoir sampling

sampling Gumbel reservoir-sampling sampling-without-replacement

A while back, Hanna and I stumbled upon the following blog post: Algorithms Every Data Scientist Should Know: Reservoir Sampling, which got us excited about reservoir sampling.

Around the same time, I attended a talk by Tamir Hazan about some of his work on perturb-and-MAP (Hazan & Jaakkola, 2012), which is inspired by the Gumbel-max-trick (see previous post). The apparent similarity between weighted reservoir sampling and the Gumbel-max trick lead us to make some cute connections, which I'll describe in this post.

The problem: We're given a stream of unnormalized probabilities, \(x_1, x_2, \cdots\). At any point in time \(t\) we'd like to have a sampled index \(i\) available, where the probability of \(i\) is given by \(\pi_t(i) = \frac{x_i}{ \sum_{j=1}^t x_j}\).

Assume, without loss of generality, that \(x_i > 0\) for all \(i\). (If any element has a zero weight we can safely ignore it since it should never be sampled.)

Streaming Gumbel-max sampler: I came up with the following algorithm, which is a simple "modification" of the Gumbel-max-trick for handling streaming data:

\(a = -\infty; b = \text{null} \ \ \text{# maximum value and index}\)
for \(i=1,2,\cdots;\) do:
# Compute log-unnormalized probabilities
\(w_i = \log(x_i)\)
# Additively perturb each weight by a Gumbel random variate
\(z_i \sim \text{Gumbel}(0,1)\)
\(k_i = w_i + z_i\)
# Keep around the largest \(k_i\) (i.e. the argmax)
if \(k_i > a\):
\(\ \ \ \ a = k_i\)
\(\ \ \ \ b = i\)

If we interrupt this algorithm at any point, we have a sample \(b\).

After convincing myself this algorithm was correct, I sat down to try to understand the algorithm in the blog post, which is due to Efraimidis and Spirakis (2005) (paywall, free summary). They looked similar in many ways but used different sorting keys/perturbations.

Efraimidis and Spirakis (2005): Here is the ES algorithm for weighted reservoir sampling

\(a = -\infty; b = \text{null}\)
for \(i=1,2,\cdots;\) do:
# compute randomized key
\(u_i \sim \text{Uniform}(0,1)\)
\(e_i = u_i^{(\frac{1}{x_i})}\)
# Keep around the largest \(e_i\)
if \(k_i > a\):
\(\ \ \ \ a = k_i\)
\(\ \ \ \ b = i\)

Again, if we interrupt this algorithm at any point, we have our sample \(b\). Note that you can simplify \(e_i\) so that you don't have to compute pow (which is nice because pow is pretty slow). It's equivalent to use \(e'_i = \log(e_i) = \log(u_i)/x_i\) because \(\log\) is monotonic. (Note that \(-e'_i \sim \textrm{Exponential}(x_i)\).)

Relationship: At a high level, we can see that both algorithms compute a randomized key and take an argmax. What's the relationship between the keys?

First, note that a \(\text{Gumbel}(0,1)\) variate can be generated via \(-\log(-\log(\text{Uniform}(0,1)))\). This is a straightforward application of the inverse transform sampling method for random number generation. This means that if we use the same sequence of uniform random variates then, \(z_i = -\log(-\log(u_i))\).

However, this does not give use equality between \(k_i\) and \(e_i\), but it does turn out that \(k_i = -\log(-\log(e_i))\), which is useful because this is a monotonic transformation on the interval \((0,1)\). Since monotonic transformations preserve ordering, the sequences \(k\) and \(e\) result in the same comparison decisions, as well as, the same argmax. In summary, the algorithms are the same!

Extensions: After reading a little further along in the ES paper, we see that the same algorithm can be used to perform sampling without replacement by sorting and taking the elements with the highest keys. This same modification applies to the Gumbel-max-trick because the keys have precisely the same ordering as ES. In practice, we don't sort the key, but instead, use a bounded priority queue.

Closing: To the best of my knowledge, the connection between the Gumbel-max trick and ES is undocumented. Furthermore, the Gumbel-max-trick is not known as a streaming algorithm, much less known to perform sampling without replacement! If you know of a reference, let me know.

How to cite this article: If you found this article useful, please cite it as

@misc{vieira2014gumbel,
    title = {Gumbel-max trick and weighted reservoir sampling},
    author = {Tim Vieira},
    url = {http://timvieira.github.io/blog/post/2014/08/01/gumbel-max-trick-and-weighted-reservoir-sampling/},
    year = {2014}
}

Further reading

I have a few other articles that are related to sampling without replacement, reservoir sampling, and Gumbel tricks.

Here are a few interesting papers that build on the ideas in this article.

Comments