search

CompEvol / beast2

Bayesian Evolutionary Analysis by Sampling Trees
www.beast2.org
GNU Lesser General Public License v2.1
240 stars 84 forks source link

BEAST should warn for variation in log frequencies #1007

Open rbouckaert opened 3 years ago

rbouckaert commented 3 years ago

Post processing programs that rely on multiple log files -- like Tracer for skyline analysis, or UglyTrees for visualising multi species coalescent trees -- rely on the log frequencies of these log files to be the same, and may fail with uninformative error messages. To prevent this happening, it may be useful to have BEAST warn about using loggers that use different log frequencies.

The screen log may be ignored, since it is not used in post-processing.

tgvaughan commented 3 years ago

Hi Remco, could we instead harden those programmes? I often have some loggers that either produce a lot of output or require a lot of computation, so having a factor of 10 difference between logging frequencies occurs very frequently in my XMLs and the XMLs produced by my BEAUti templates. Having these default configurations display scary warning messages would be a shame.

rbouckaert commented 3 years ago

Hi Tim, good point. One way to reduce log sizes is to use fewer significant digits in real values that are logged (there is a dp input on TreeWithMetaDataLogger that takes care of that for trees, for example), which can reduce log sizes considerably. In the recent past I encountered three cases where the post-processor got confused by differences in log frequencies -- twice with Tracer and once with UglyTrees -- which takes up unnecessary time to trace back to the origin. I believe some work is underway to make Tracer produce more sensible error messages when trace and tree logs are of different frequencies, and talked with Jordan about putting a warning in UglyTrees, but suspect there are more post-processing tools out there that are out of our control.

How about producing a non-scary warning message: currently, I have

WARNING: Loggers appear to have different log frequency.
WARNING: This may cause problems in post-processing based on more than one log files.
WARNING: Therefore, it is recommended to use the same log frequency

perhaps change this to:

WARNING: Loggers appear to have different log frequency, which may be perfectly fine, but may cause problems in post-processing based on multiple log files. Consider using the same log frequency if you plan to do demographic reconstruction or visualising gene trees with species trees.

Happy to hear any other suggestion.

tgvaughan commented 2 years ago

Hi Remco, sorry for the long delay in getting back. To be clear, I'm not talking about "regular" loggers here, but ones that include some non-trivial state augmentation. Think stochastic mapping of traits and/or trajectories. These aren't just producing large files, but also involve a lot of computation.

I'm happy with the updated warning message. But, if they're reading this, I also would really encourage those responsible for the failing post-processing software to fix these too - the log files have the iteration numbers embedded in them, so it should be easy for then to filter out the additional lines from the higher-frequency file.