Bug #1439

ctobssim throws incomprehensible exception when using a binned observation as input

Added by Mayer Michael over 11 years ago. Updated over 11 years ago.

Status:ClosedStart date:03/10/2015
Priority:NormalDue date:
Assigned To:Mayer Michael% Done:

100%

Category:-Estimated time:3.00 hours
Target version:1.0.0
Duration:

Description

Currently, ctobssim is not intended to be used with a binned observation as input. Another tool is planned to achieve that (#544).

Forgetting about this rule and running ctobssim with a count map as input, the error message does not reflect the actual error and appears cryptic. This should be changed.
Here is my command line output:

mimayer ~/Software/ctools $ ctobssim inobs=test/data/crab_cntmap.fits.gz 
Calibration database [dummy] 
Instrument response function [cta_dummy_irf] 
Model [$CTOOLS/share/models/crab.xml] 
Output event data file or observation definition file [events.fits] 
*** ERROR encounterted in the execution of ctobssim. Run aborted ...
*** ERROR in GCTAEventCube::naxis(int): CTA event cube axis 208182906 is outside the valid range [0,2].

Recurrence

No recurrence.

History

#1 Updated by Knödlseder Jürgen over 11 years ago

  • Target version set to 1.0.0

#2 Updated by Knödlseder Jürgen over 11 years ago

  • Project changed from GammaLib to ctools
  • Target version deleted (1.0.0)

#3 Updated by Knödlseder Jürgen over 11 years ago

  • Target version set to 1.0.0

And check this for all other tools!

#4 Updated by Mayer Michael over 11 years ago

  • Status changed from New to In Progress
  • Assigned To set to Mayer Michael

#5 Updated by Mayer Michael over 11 years ago

I think the ctools and cscripts can be separated into two kinds. There are tools that require unbinned observation as input and some that should be able to run on both (binned and unbinned):

Unbinned only:
  • ctselect
  • ctobssim
  • ctexpcube
  • ctpsfcube
  • ctbckcube
  • ctbin
  • ctskymap
  • csspec
  • cslightcrv
  • cspull
  • cssens
  • cstsdist
Both binned and unbinned:
  • ctbutterfly
  • ctcubemask (binned only?)
  • cterror
  • ctlike
  • ctmodel
  • cttsmap
  • ctulimit
  • csresmap

We should verify that each tool is able to run in the modes listed above. In addition, the error message for a wrong input observation type should be changed to be reasonably understood by a user.

#6 Updated by Knödlseder Jürgen over 11 years ago

Mayer Michael wrote:

  • ctcubemask (binned only?)

Yes

#7 Updated by Mayer Michael over 11 years ago

We decided to simply skip binned observations in tools where unbinned observations are necessary and write that as warning into the logger.
This has now been implemented for the following tools:
  • ctselect
  • ctobssim
  • ctpsfcube
  • ctexpcube
  • ctbkgcube
  • ctskymap
  • ctbin

The corresponding cscripts don’t have to be changed as they just call some of the ctools above.
The tool ctcubemask is fine already as it skips unbinned observations already.
In addition, tools working on the likelihood profile also seem to be alright in this matter as they can take both inputs.

Two tools are a bit different to handle:
  1. ctmodel: First, I noticed that there is a get_obs() function which basically does the same as ctool::get_observations(). Therefore, I removed this function and use the generic ctool function instead. The other question is if we really want ctmodel to work on binned observations, which it currently does. However, it would then not be aligned with ctbin, which only creates a cube from the unbinned observations. Therefore the creation of a residual map could become a pain. If ctmodel just adds up all the model counts including binned observation (which is not considered in ctbin), the diagnostic might become tricky.
  2. csresmap: This tool does not function for a binned input, as it runs ctbin and ctmodel. We could check for a binned observation and, if present, neglect the ctbin step. Of course, this strongly depends on how we decide that ctmodel should behave.

An idea for ctmodel would be to collapse all available unbinned observations into one binned observation but creates one model cube per given binned observation.

The updated code is available in
  • gammalib on 1439-change-usage-of-binned-observations-in-GCTACube-classes
  • ctools on 1439-adapt-ctools-to-consistently-handle-observation-types

#8 Updated by Mayer Michael over 11 years ago

  • Status changed from In Progress to Feedback
  • % Done changed from 0 to 90
  • Estimated time set to 3.00

#9 Updated by Knödlseder Jürgen over 11 years ago

Mayer Michael wrote:

Two tools are a bit different to handle:
  1. ctmodel: First, I noticed that there is a get_obs() function which basically does the same as ctool::get_observations(). Therefore, I removed this function and use the generic ctool function instead. The other question is if we really want ctmodel to work on binned observations, which it currently does. However, it would then not be aligned with ctbin, which only creates a cube from the unbinned observations. Therefore the creation of a residual map could become a pain. If ctmodel just adds up all the model counts including binned observation (which is not considered in ctbin), the diagnostic might become tricky.

It would be nice to have ctmodel also working for binned observation, but I understand your point of inconsistent behavior. What about the following: if ctmodel gets an observation definition file or an event list it will work in the mode where it only uses the unbinned data. If however a counts cube is provided on input, it will compute a model for exactly that cube.

I was wondering whether it would make sense to rename the tool to fit the names of the other tools, e.g., ctmodcube? This also raises the question whether ctbin should in fact be renamed. ctbin has the “historic” name from the Fermi Science Tools, but to fit in the suites of the other tools one could think about calling that tool ctevtcube or ctcntcube. Maybe something that we should discuss all together.

  1. csresmap: This tool does not function for a binned input, as it runs ctbin and ctmodel. We could check for a binned observation and, if present, neglect the ctbin step. Of course, this strongly depends on how we decide that ctmodel should behave.
Would it make sense to do the following:
  • if an observation definition file is provided, use only the unbinned observations in a way similar to ctbin, etc.
  • if an event list is provided, to as for observation definition file
  • if a counts cube is provided, skip the ctbin step and use it directly in ctmodel

#10 Updated by Mayer Michael over 11 years ago

  • Status changed from Feedback to Pull request
  • % Done changed from 90 to 100

I made the proposed changes on the branches listed above.
For csresmap, I made the parameter algorithm unhidden and further allowed to two hidden parameters: cntcube and modcube. If the user specifies those two on construction, the tool will automatically skip ctbin and ctmodel and compute the residual directly (similar to farith).

#11 Updated by Knödlseder Jürgen over 11 years ago

Mayer Michael wrote:

  • gammalib on 1439-change-usage-of-binned-observations-in-GCTACube-classes
  • ctools on 1439-adapt-ctools-to-consistently-handle-observation-types

I did the 1439-change-usage-of-binned-observations-in-GCTACube-classes branch, enhancing a bit the code so that also information about non-CTA observations is logged.

I will now proceed with the ctools branch.

#12 Updated by Knödlseder Jürgen over 11 years ago

I homogenized the output and tested the following tools with success:
  • ctobssim
  • ctselect
  • ctbin
  • ctexpcube
  • ctpsfcube
  • ctbkgcube
  • ctskymap

#13 Updated by Knödlseder Jürgen over 11 years ago

I also checked ctmodel and removed the cntcube parameter from csresmap. If a counts cube is specified as inobs, the script will ask for the modcube parameter first. If it is given it simply computes the residual, otherwise it will continue querying the response parameters.

#14 Updated by Knödlseder Jürgen over 11 years ago

  • Status changed from Pull request to Feedback

I think we can close this now.

#15 Updated by Mayer Michael over 11 years ago

I agree. I will probably run some further checks next week. But any occurring problem should then result in a new issue.

#16 Updated by Knödlseder Jürgen over 11 years ago

  • Status changed from Feedback to Closed

Also available in: Atom PDF