How to set the "Burnin" in LogCombiner

[Chogo]

I though i would post this as i had some trouble trying for figure
this out a while back

So you have your XML file which contains your Chain Length (C) and
your Sampling period/"logevery" (L)

for example my chain length could be C=50,000,000 and i choose to log
every L=10,000.

After doing multiple runs (ensuring the taxon order remains the same),
you might want to combine your .log and .tree files.

People often get confused when setting the burnin, as it is different
whether you are combining log files or tree files. What you first do
is pick your burnin percentage (B). The standard would be B=0.1 or
10%.

For the LOG FILE - the number you put into the burnin box will be ----
> C*B. Or in my case 50,000,000*0.1 = 5,000,000

For the TREE FILE - number you put into the burnin box will be ----->
(C/L)*B. Or in my case (50,000,000/10,000)*0.1 = 500

Hope this makes sense and can provide some help

[Simey]

Thank you Chogo! Very useful information.

[Satish Gaikwad]

Indeed very useful information. since many days I was stuck at this point. Thank you.

[Jarmila Smiešková]

This seems to be clear to me, but I have another question.

Firstly, you set burnin for trees in LogCombiner (e.g, 10%), but secondly, what should I set as a burnin in TreeAnnotator?

Many thanx for reactions.

Dne pondělí, 17. ledna 2011 15:45:41 UTC+1 Chogo napsal(a):

[Natalia]

Hello Jarmila, 

If you have set a burn in for each file in log combiner, then you don't need a burn in for the combined file when you use tree annotator, because the burn in has already been removed by log combiner.

However, if you are using tree annotator on a file with burn in, you should set the burn in in terms of number of samples, not mcmc generations (see the first post in this thread).

I hope this helps,

Natalia

[Jarmila Smiešková]

Thank you very much, Natalia! Very useful information.

Dne pátek, 28. března 2014 16:57:27 UTC+1 Natalia napsal(a):

[Michael Forthman]

This is helpful for me when I'm dealing with one log file only, however, I'm trying to combine multiple log files. When I set the burn in, my combined files has a total number of samples that do not equal to the number it should be. For example, I have 200,000,000 samples/states from each run. I set burn-in to 20,000,000 for each log file in Tracer. If I combined the log files in LogCombiner with 10% burn-in, I get a different number of combined samples than what I get in Tracer, yet still not matching the number it really should be. If I combine the log files in LogCombiner without indicating burn-in, open it in Tracer, and specify 10% burn-in, I get a similar situation: the number of combined samples is not consistent with the other approaches and is still less than what it should be.

[Hu Ling]

I checked the log files. Assuming I run the same XML file twice, with MCMC set to 10000 and sampling 100, the MCMC starts from state 0. That is, the states are 0, 100, 200, and so on up to 10000. So the actual output should have 10000/100 + 1 = 101 rows. However, LogCombiner seems to include state 0 in the count. When burn-in is set to 0.1 (10%), the resulting merged file ends up with an extra two samples, calculated as 10000 × 0.1 + 100 × 2 = 1100. This is my current understanding. I'm not sure if it's correct.