Prior plots and shuffle

[MattB]

Hello,

The prior curves for some of the parameters are not a flat line
(almost flat but slightly wavy) as I would expect for a uniform
distribution. Is this a sign of bias in the randomization?

Secondly, I am interested in understanding the contents of INTFILE and
how shuffle.exe works. What does the #iterations command-line
argument do?

Thanks!
Matt

[Joao Sollari Lopes]

Hi Matt,

Sorry for the late reply.

Regarding the prior curves, the R scripts is reading the .pri file,
which has random samples from the prior, then, using the library
locfit, it fits a curve on the distribution of points. This will give
you an idea of the distribution of the prior, but it won't give you
the exact distribution you used. For a better inspection by-eye try
plotting an histogram of the prior instead.

Note that, for graphical purposes, you can use the following command
for uniform distributions:
curve(dunif(x,min, max),min,max)

As for the random number generator, INTFILE is a list of 99 numbers:
98 random numbers and a number 'i' that goes from 0 to 97. Whenever
popabc needs to obtain a random number it picks the ith number from
the list of random numbers and replace it with the result of that
number to the power of the random number in the list that is 27
positions forward. Then the number 'i' is incremented one unit.
What shuffle does is to perform this operation 'x' number of times,
where 'x' is the argument given in the command line.

If you're starting two runs with the exact some conditions using the
exact same INTFILE it will give you the exact same results. If you use
shuffle.exe the INTFILE will be "at a different time", thus, giving
you different results. You should use shuffle if you want to run
several jobs with the same conditions.

Hope that makes it clearer.

Best,
Joao

[MattB]

Very helpful, thank you!

On Apr 13, 2:54 am, Joao Sollari Lopes <j.sollari.lo...@gmail.com>
wrote: