Archive for William Feller

easy riddle

Posted in Books, Kids, R with tags , , , , , on July 12, 2017 by xi'an

From the current Riddler, a problem that only requires a few lines of code and a few seconds of reasoning. Or not.

N households each stole the earnings from one of the (N-1) other households, one at a time. What is the probability that a given household is not burglarised? And what are the expected final earnings of each household in the list, assuming they all start with $1?

The first question is close to Feller’s enveloppe problem in that


is close to exp(-1) for N large. The second question can easily be solved by an R code like

for (v in 1:M){
 while (min(abs(vole-(1:N)))==0)
 for (t in 1:N){

which returns a pretty regular exponential-like curve, although I cannot figure the exact curve beyond the third burglary. The published solution gives the curve

{\frac{N-2}{N-1}}^{999}\times 2+{\frac{1}{N-1}}^{t-1}\times{\frac{N-1}{N}}^{N-t}\times\frac{N}{N-1}

corresponding to the probability of never being robbed (and getting on average an extra unit from the robbery) and of being robbed only before robbing someone else (with average wealth N/(N-1)).

sampling by exhaustion

Posted in Books, Kids, R, Statistics with tags , , , , on November 25, 2016 by xi'an

The riddle set by The Riddler of last week sums up as follows:

Within a population of size N, each individual in the population independently selects another individual. All individuals selected at least once are removed and the process iterates until one or zero individual is left. What is the probability that there is zero individual left?

While I cannot see a clean analytical solution to this problem, it reminds me of an enveloppe-versus-letter (matches) problem I saw in graduate school. Indeed, the expected number of removed (or selected) individuals is given by


which is equivalent to (1-e⁻¹)N for N large, meaning that the population decreases by an average factor of e⁻¹ at each round. And that it takes on average approximately log(N) iterations to reach a single individual. A simulation of the first probabilities of ending with one individual led me to the above curve, which wiggles in an almost periodic way around the probability ½, equal to the average of those probabilities. Using the R code

rad=function(N){#next population size
  for (i in 2:N)#sampling
for (N in 3:M){
 for (t in 1:T){#one single step
   if (i>0) prop=prop+sal[i]}

which exploits the previously computed probabilities. The variability is most interesting if unexpected, but looking back at Feller‘s sections and exercises on the classical occupancy problem, I could not find a connection with this problem. If it exists. Still, if N is large enough, the exclusion of one index from the selection becomes negligible and the probability of moving from n to m individuals should be approximately [Feller, eqn (2.4), p.102]

p_n(m)={n\choose m}\sum_{v=}^{n-m} (-1)^v {n-m\choose v} \left(1-\frac{m+v}{n}\right)^n

This formula approximates quite well the exact probability, but as in a previous post about the birthday problem, it proves quite delicate to compute. As already noticed by Feller.

random walk on a torus [riddle]

Posted in Books, Kids, pictures with tags , , , , , , , , , on September 16, 2016 by xi'an

Galgate, Lancastershire, July 19, 2011The Riddler of this week(-end) has a simple riddle to propose, namely given a random walk on the {1,2,…,N} torus with a ⅓ probability of death, what is the probability of death occurring at the starting point?

The question is close to William Feller’s famous Chapter III on random walks. With his equally famous reflection principle. Conditioning on the time n of death, which as we all know is definitely absorbing (!), the event of interest is a passage at zero, or any multiple of N (omitting the torus cancellation), at time n-1 (since death occurs the next time). For a passage in zero, this does not happen if n is even (since n-1 is odd) and else it is a Binomial event with probability

{n \choose \frac{n-1}{2}} 2^{-n}

For a passage in kN, with k different from zero, kN+n must be odd and the probability is then

{n \choose \frac{n-1+kN}{2}} 2^{-n}

which leads to a global probability of

\sum_{n=0}^\infty \dfrac{2^n}{3^{n+1}} \sum_{k=-\lfloor (n-1)/N \rfloor}^{\lfloor (n+1)/N \rfloor} {n \choose \frac{n-1+kN}{2}} 2^{-n}


\sum_{n=0}^\infty \dfrac{1}{3^{n+1}} \sum_{k=-\lfloor (n-1)/N \rfloor}^{\lfloor (n+1)/N \rfloor} {n \choose \frac{n-1+kN}{2}}

Since this formula is rather unwieldy I looked for another approach in a métro ride [to downtown Paris to enjoy a drink with Stephen Stiegler]. An easier one is to allocate to each point on the torus a probability p[i] to die at position 1 and to solve the system of equations that is associated with it. For instance, when N=3, the system of equations is reduced to

p_0=1/3+2/3 p_1, \quad p_1=1/3 p_0 + 1/3 p_1

which leads to a probability of ½ to die at position 0 when leaving from 0. When letting N grows to infinity, the torus structure no longer matters and the probability of dying at position 0 implies returning in position 0, which is a special case of the above combinatoric formula, namely

\sum_{m=0}^\infty \dfrac{1}{3^{2m+1}}  {2m \choose m}

which happens to be equal to

\dfrac{1}{3}\,\dfrac{1}{\sqrt{1-4/9}}=\dfrac{1}{\sqrt{5}}\approx 0.4472

as can be [unnecessarily] checked by a direct R simulation. This √5 is actually the most surprising part of the exercise!

another riddle

Posted in Books, Kids, Statistics with tags , , , on August 9, 2016 by xi'an

A riddle from The Riddler, a few weeks ago, to ponder while waiting for Godot another plane:

Consider a game played with a coin, between you and a friend, on the integer line. You are assigned a negative integer -X, and your friend is assigned a positive integer, +Y. A marker is placed at zero on the number line. Then the coin is repeatedly flipped. Every time the coin lands heads, the marker is moved one integer in a positive direction. Every time the coin lands tails, the marker moves one integer in a negative direction. You win if the coin reaches -X first, while your friend wins if the coin reaches +Y first.

What is the expected number of coin flips in a complete game?

Not much to say there since this is the classical Feller’s gambler ruin problem that I used to analyse in my practical classes at Polytechnique. The winning probability is X/(X+Y). Assuming the coin is fair. And the distribution of the hitting times (win or loose) is easily analysed by the reflection principle. But a simpler recursion shows that the expected number of steps till hitting one boundary is XY, which is rather surprising as a multiplicative formula.

the Flatland paradox [#2]

Posted in Books, Kids, R, Statistics, University life with tags , , , , , , , , , , , on May 27, 2015 by xi'an

flatlandAnother trip in the métro today (to work with Pierre Jacob and Lawrence Murray in a Paris Anticafé!, as the University was closed) led me to infer—warning!, this is not the exact distribution!—the distribution of x, namely

f(x|N) = \frac{4^p}{4^{\ell+2p}} {\ell+p \choose p}\,\mathbb{I}_{N=\ell+2p}

since a path x of length l(x) will corresponds to N draws if N-l(x) is an even integer 2p and p undistinguishable annihilations in 4 possible directions have to be distributed over l(x)+1 possible locations, with Feller’s number of distinguishable distributions as a result. With a prior π(N)=1/N on N, hence on p, the posterior on p is given by

\pi(p|x) \propto 4^{-p} {\ell+p \choose p} \frac{1}{\ell+2p}

Now, given N and  x, the probability of no annihilation on the last round is 1 when l(x)=N and in general

\frac{4^p}{4^{\ell+2p}}{\ell-1+p \choose p}\big/\frac{4^p}{4^{\ell+2p}}{\ell+p \choose p}=\frac{\ell}{\ell+p}=\frac{2\ell}{N+\ell}

which can be integrated against the posterior. The numerical expectation is represented for a range of values of l(x) in the above graph. Interestingly, the posterior probability is constant for l(x) large  and equal to 0.8125 under a flat prior over N.

flatelGetting back to Pierre Druilhet’s approach, he sets a flat prior on the length of the path θ and from there derives that the probability of annihilation is about 3/4. However, “the uniform prior on the paths of lengths lower or equal to M” used for this derivation which gives a probability of length l proportional to 3l is quite different from the distribution of l(θ) given a number of draws N. Which as shown above looks much more like a Binomial B(N,1/2).

flatpostHowever, being not quite certain about the reasoning involving Fieller’s trick, I ran an ABC experiment under a flat prior restricted to (l(x),4l(x)) and got the above, where the histogram is for a posterior sample associated with l(x)=195 and the gold curve is the potential posterior. Since ABC is exact in this case (i.e., I only picked N’s for which l(x)=195), ABC is not to blame for the discrepancy! I asked about the distribution on Stack Exchange maths forum (and a few colleagues here as well) but got no reply so far… Here is the R code that goes with the ABC implementation:

#ABC version
for (t in 1:T){
#generate a path
#eliminate U-turns
  while (sum(uturn>0)){
#subsample to get exact posterior

the Flatland paradox

Posted in Books, Kids, R, Statistics, University life with tags , , , , , , , , on May 13, 2015 by xi'an

Pierre Druilhet arXived a note a few days ago about the Flatland paradox (due to Stone, 1976) and his arguments against the flat prior. The paradox in this highly artificial setting is as follows:  Consider a sequence θ of N independent draws from {a,b,1/a,1/b} such that

  1. N and θ are unknown;
  2. a draw followed by its inverse and this inverse are removed from θ;
  3. the successor x of θ is observed, meaning an extra draw is made and the above rule applied.

Then the frequentist probability that x is longer than θ given θ is at least 3/4—at least because θ could be zero—while the posterior probability that x is longer than θ given x is 1/4 under the flat prior over θ. Paradox that 3/4 and 1/4 clash. Not so much of a paradox because there is no joint probability distribution over (x,θ).

The paradox was actually discussed at length in Larry Wasserman’s now defunct Normal Variate. From which I borrowed Larry’s graphical representation of the four possible values of θ given the (green) endpoint of x. Larry uses the Flatland paradox hammer to fix another nail on the coffin he contemplates for improper priors. And all things Bayes. Pierre (like others before him) argues against the flat prior on θ and shows that a flat prior on the length of θ leads to recover 3/4 as the posterior probability that x is longer than θ.

As I was reading the paper in the métro yesterday morning, I became less and less satisfied with the whole analysis of the problem in that I could not perceive θ as a parameter of the model. While this may sound a pedantic distinction, θ is a latent variable (or a random effect) associated with x in a model where the only unknown parameter is N, the total number of draws used to produce θ and x. The distributions of both θ and x are entirely determined by N. (In that sense, the flatland paradox can be seen as a marginalisation paradox in that an improper prior on N cannot be interpreted as projecting a prior on θ.) Given N, the distribution of x of length l(x) is then 1/4N times the number of ways of picking (N-l(x)) annihilation steps among N. Using a prior on N like 1/N , which is improper, then leads to favour the shortest path as well. (After discussing the issue with Pierre Druilhet, I realised he had a similar perspective on the issue. Except that he puts a flat prior on the length l(x).) Looking a wee bit further for references, I also found that Bruce Hill had adopted the same perspective of a prior on N.

Feller’s shoes and Rasmus’ socks [well, Karl’s actually…]

Posted in Books, Kids, R, Statistics, University life with tags , , , , on October 24, 2014 by xi'an

Yesterday, Rasmus Bååth [of puppies’ fame!] posted a very nice blog using ABC to derive the posterior distribution of the total number of socks in the laundry when only pulling out orphan socks and no pair at all in the first eleven draws. Maybe not the most pressing issue for Bayesian inference in the era of Big data but still a challenge of sorts!

Rasmus set a prior on the total number m of socks, a negative Binomial Neg(15,1/3) distribution, and another prior of the proportion of socks that come by pairs, a Beta B(15,2) distribution, then simulated pseudo-data by picking eleven socks at random, and at last applied ABC (in Rubin’s 1984 sense) by waiting for the observed event, i.e. only orphans and no pair [of socks]. Brilliant!

The overall simplicity of the problem set me wondering about an alternative solution using the likelihood. Cannot be that hard, can it?! After a few computations rejected by opposing them to experimental frequencies, I put the problem on hold until I was back home and with access to my Feller volume 1, one of the few [math] books I keep at home… As I was convinced one of the exercises in Chapter II would cover this case. After checking, I found a partial solution, namely Exercice 26:

A closet contains n pairs of shoes. If 2r shoes are chosen at random (with 2r<n), what is the probability that there will be (a) no complete pair, (b) exactly one complete pair, (c) exactly two complete pairs among them?

This is not exactly a solution, but rather a problem, however it leads to the value


as the probability of obtaining j pairs among those 2r shoes. Which also works for an odd number t of shoes:


as I checked against my large simulations. socksSo I solved Exercise 26 in Feller volume 1 (!), but not Rasmus’ problem, since there are those orphan socks on top of the pairs. If one draws 11 socks out of m socks made of f orphans and g pairs, with f+2g=m, the number k of socks from the orphan group is an hypergeometric H(11,m,f) rv and the probability to observe 11 orphan socks total (either from the orphan or from the paired groups) is thus the marginal over all possible values of k:

\sum_{k=0}^{11} \dfrac{\binom{f}{k}\binom{2g}{11-k}}{\binom{m}{11}}\times\dfrac{2^{11-k}\binom{g}{11-k}}{\binom{2g}{11-k}}

so it could be argued that we are facing a closed-form likelihood problem. Even though it presumably took me longer to achieve this formula than for Rasmus to run his exact ABC code!