Uniforms summing to a uniform

A code golf question by xnor led to the following nice problem: let X and Y be 2 random variables such that marginally, X\sim U(0,1) and Y\sim U(0,1). Find a joint distribution of (X,Y) such that X+Y\sim U(\frac12, \frac32).

You need X and Y to be negatively correlated for this. I wrote the problem in the lab coffee room, leading to nice discussions (see also Xian’s blog post). Here are two solutions to the problem:

1. Let X\sim U(0,1) and Y = \begin{cases}1-2X\text{ if }X<\frac12\\2-2X\text{ if } X\geq\frac12\end{cases}.  Then:

  • Y|X<\frac12 and Y|X\geq\frac12 are both U(0,1), hence Y\sim U(0,1)
  • X+Y|X<\frac12 = 1-X|X<\frac12\sim U(\frac12, 1) and X+Y|X\geq\frac12 = 2-X|X\geq\frac12\sim U(1, \frac32) hence X+Y\sim U(\frac12, \frac32)

2. A second solution, found by my colleague Amic Frouvelle, is to sample (X,Y) uniformly from the black area:

cuef8-1

I quite like that the first solution is 1d but the second is 2d.

 

On Unbiased MCMC with couplings

Pierre Jacob, John O’Leary and Yves Atchadé’s excellent paper on Unbiased MCMC with couplings will be read at the Royal Statistical Society tomorrow; Pierre has already presented the paper on the Statisfaction blog.
Although we won’t be present tomorrow, we have read it at length in our local reading group with Xian Robert and PhD students Grégoire Clarté, Adrien Hairault and Caroline Lawless, and have submitted the following discussion.

We congratulate the authors for this excellent paper.

In “traditional” MCMC, it is standard to check that stationarity has been attained by running a small number of parallel chains, initiated at different starting points, to verify that the final distribution is independent of the initialization — even though the single versus multiple chain(s) debate errupted from the start with Gelman and Rubin (1992) versus Geyer (1992).

As noted by the authors, a bad choice of the initial distribution p_0 can lead to poor properties. In essence, this occurs and remains undetected for the current proposal because the coupling of the chains occurs long before the chain reaches stationarity. We would like to make two suggestions to alleviate this issue, and hence add a stationarity check as a byproduct of the run.

  1. The chains X and Y need to have the same initial distribution, but different pairs of chains on different parallel cores can afford different initial distributions. The resulting estimator remains unbiased. We would therefore suggest that parallel chains be initiated from distributions which put weight on different parts of the parameter space. Ideas from the Quasi-Monte Carlo literature (see Gerber & Chopin 2015) could be used here.
  2.  We also note that although the marginal distributions of X and Y need to be identical, any joint distribution on (X,Y) produces an unbiased algorithm. We would suggest that it is preferable that X and Y meet (shortly) after the chains have reached stationarity. Here is one possible strategy to this end: let p and p' be two distributions which put weight on different parts of the space, and Z\sim Bernoulli(1/2). If Z=0, take X_0\sim p and Y_0\sim p', else take X_0\sim p' and Y_0\sim p. The marginal distribution of both X_0 and Y_0 is \frac12(p+p'), but the two chains will start in different parts of the parameter space and are likely to meet after they have both reached stationarity.

The ideal algorithm is one which gives a correct answer when it has converged, and a warning or error when it hasn’t. MCMC chains which have not yet reached stationarity (for example because they have not found all modes of a multimodal distribution) can be hard to detect. Here, this issue is more likely to be detected since it would lead to the coupling not occuring: \mathbb E[\tau] is large, and this is a feature, since it warns the practitioner that their kernel is ill-fitted to the target density.

Black on black tooltips in Firefox with Kubuntu

I use Firefox on Kubuntu, and for a long time I had an issue with the tooltips: the characters were printed in black on a black background (a slightly different shade of black, but still very difficult to read).

I used to have a solution with Stylish, but it broke in Firefox 57 (Firefox Quantum). Here is a solution which works now, for anyone else with the same issue.

  • Navigate to ~/.mozilla/firefox/
  • Find your Firefox profile: a folder with a name like 1rsnaite.default
  • Navigate to ~/.mozilla/firefox/1rsnaite.default/chrome/ or whatnot (you might need to create the chrome/ folder)
  • Using your favourite text editor, open the file ~/.mozilla/firefox/1rsnaite.default/chrome/userChrome.css (creating it if necessary)
  • In this file, put the following code:
  • /* AGENT_SHEET */
    
    @namespace xul url(http://www.mozilla.org/keymaster/gatekeeper/there.is.only.xul);
    
    #btTooltip,
    #un-toolbar-tooltip,
    #tooltip,
    .tooltip,
    #aHTMLTooltip,
    #urlTooltip,
    tooltip,
    #aHTMLTooltip,
    #urlTooltip,
    #brief-tooltip,
    #btTooltipTextBox,
    #un-toolbar-tooltip
    {
     color:#FFFFFF !important;
    }
  • Save and restart Firefox.
  • If you have several profiles, repeat for the other profiles.

I am not an expert at these things; if this does not work for you, I won’t be able to help you any better than Google.

I used the following sites to find this solution:

Lecturer position in Statistics at Dauphine

An associate professor (“Maître de conférences”) position in Applied or Computational Statistics is expected to be open at Université Paris-Dauphine. The recruitment process will mostly take place during the spring, for an appointment date of 1 September 2017.

However, candidates must first go through the national “qualification”. This process should not be problematic, but is held much earlier in the year: you need to sign up by 25 October (next week!), then send some documents by December. Unfortunately, the committee cannot consider applications from candidates who do not hold the “qualification”.

If you need help with the process, feel free to contact me.

David Cox is the inaugural recipient of the International Prize in Statistics

David Cox was announced today as the inaugural recipient of the International Prize in Statistics.

My first foray into Statistics was an analysis of Cox models I did for my undegraduate thesis at ENS in 2005. I had no idea back then that David Cox was still alive and active; in my mind, he was a historical figure, on par with other great mathematicians who gave their names to objects of study — Euler, Galois, Lebesgue…

When I arrived at Oxford a few months later, I was amazed to meet him, and to see that he was still very active, both as a researcher and as the organizer of events for doctoral students.

David Cox is the perfect choice as the first person to receive this prize. I hope that the inauguration of this prize will help show the public that Statistics require complex and innovative methods, that have been tackled by some exceptional minds, and should not be seen as a “sub-science” compared to other more “noble” sciences.

MCMSki 4

I am attending the MCMSki 4 conference for the next 3 days; I guess I’ll see many of you there!

I am organizing a session on Wednesday morning on Advances in Monte Carlo motivated by applications; I’m looking forward to hearing the talks of Alexis Muir-Watt, Simon Barthelmé, Lawrence Murray and Rémi Bardenet during that session, as well as the rest of the very strong programme.

I’ll also be part of the jury for the best poster prize; there are many promising abstracts.