Where should the settings for Parameter Estimation be determined?
Currently, most of the settings of Parameter Estimation such as seglen
, fhigh
, etc. are determined in GWCelery. It is cleaner in the sense that all the settings are visible in GWCelery and are determined in one place. However, I think it will be better to put part of the decisions (seglen
, fhigh
, srate
, distance-max
, psd-start-time
and psd-length
) in lalinference_pipe
. There are a few reasons for that.
- Removing the codes to determine these settings in
lalinference_pipe
will makelalinference_pipe
user-unfriendly even if ini file is uploaded to every superevent. For example, some people may want to launch runs with differentapprox
orengine
. In this case, just specifying GID and runninglalinference_pipe
with use-specifiedapprox
andengine
will be easier than downloading ini file from GraceDB, changing all the executable paths, changing types and channels and modifying ini file as users like. Also, it will make it easier to follow up events which do not have ini files such as events in O1 and O2. - We need
seglen
to determinepsd-start-time
andpsd-length
, but actualseglen
will be overwritten bylalinference_pipe
ifroq
option is True. Therefore, determining them inlalinference_pipe
will be cleaner. - It will be helpful for
lalinference_pipe
users to put functionalities to decidepsd-start-time
andpsd-length
depending onGDS-CALIB_STATE_VECTOR
inlalinference_pipe
. Some users might want to use BitsGWCelery
does not.
So my suggestion is basically to put decision making processes except for https://git.ligo.org/emfollow/gwcelery/blob/master/gwcelery/templates/online_pe.ini#L4-24 in lalinference_pipe
.
I think splitting these decision making processes into GWCelery
and lalinference_pipe
makes sense since seglen
, fhigh
, srate
, distance-max
, psd-start-time
and psd-length
are basically not use-specified values but should be automatically determined by source masses and detector state, while approx
, engine
, nparallel
and neff
are usually user-specified values.
cc: @vivien