The two translators of our book in Japanese, Kazue & Motohiro Ishida, contacted me about some R code mistakes in the book. The translation is nearly done and they checked every piece of code in the book, an endeavour for which I am very grateful! Here are the two issues they have noticed (after incorporating the typos signaled in the overall up-to-date summary):
First, in Example 4.4, I omitted some checkings and forgot about a minus sign, meaning Figure 4.4 (right) is wrong. (The more frustrating since this example covers perplexity!) The zeros must be controlled via code lines like
> wachd[wachd<10^(-10)]=10^(-10)
instead of the meaningless
wachd[apply(wachd,2,cumsum)<10^(-10)]=10^(-10)
and the addition of
> plex[plex>0]=0 > plech[plech>0]=0
after the definition of those two variables. (Because entropies are necessarily positive.) The most glaring omission is however the minus in
> plob=apply(exp(-plex),1,quantile,c(.025,.975)) > ploch=apply(exp(-plech),1,quantile,c(.025,.975))
which modifies Figure 4.4 in the following
The second case is Example 7.3 where I forgot to account for the log-transform of the data, which should read (p.204):
> x=c(91,504,557,609,693,727,764,803,857,929,970,1043, + 1089,1195,1384,1713) > x=log(x)
and compounded my mistake by including log-transforms of the parameters that should not be there (pp.204-205)! So (for my simulations) the posterior means of θ and σ² are 6.62 and 0.661, respectively, leading to an estimate of σ of 0.802. There should be no log transform in Exercise 7.3 either.
The same corrections apply to the French translation, most obviously…