GammaLib

Created a dedicated branch and pushed to it a version of the code that compiles (on my machine). Still remains to be tested through python scripts. Also, the branch forks a bit behind the current state of the code. Will do the merging later.

The likelihood function is at the observation level, and the call of the optimizer is done at the container level (now the standard also for the rest of the gammalib). The likelihood function slightly differs in its arguments from the other likelihood functions. Can homogeneize that later if deemed necessary.

No unit test written so far. I will first try to play with this new functionality.

Here is a script to test the ON-OFF fitting (and associated functionalities by the way). The script simulates two wobble observations with one ON region and any number of OFF regions distributed around the centre of each pointing (rough method). It allows to compare ON-OFF analysis and unbinned analysis over the entire field of view. So far, it runs and tells that no fit is performed for the ON-OFF analysis. Problem of singular matrix. Inspection under way...

Attached a new version of the test script for the ON-OFF analysis. Corrected a few bugs and added many prints for checking. The gammalib code was also modified but currently yields a segmentation fault. Hopefully this will be solved during the second coding sprint...

Latest version of the script.
A few notes:
- OFF region creation is using a rough method that works only close to equator
- Python interface at the moment does not allow using derived classes specific methods for GSkyRegion objects
- Hard-coded path to data base
- ON-OFF fit not performed properly (gammalib problem, not the script)

Revised version of the script after modifications by Anneli. She has tested the save methods for the ON-OFF data (and fixed some bugs).

Is there any stuff that is sufficiently tested so that can it be merged into devel?

Attached two scripts:
1- test_OnOff.py for testing the ON-OFF analysis on a single run: observations are simulated for two wobble pointings on either side of a point source with power-law spectrum; ON-OFF analysis is setup and executed; unbinned analysis is performed for comparison
2- pull_OnOff.py to perform a series of such runs and plot pull distributions
Both scripts are self-contained.

...during 5th coding sprint

I started with integrating the code.

First I went over the code and adapt it to the coding conventions. I also needed to adjust the so far existing unit test at it was broken due to the additional of the models parameter in the compute_response() method.

I also removed the GCTAOnOffObservations class as it is no longer needed. Now GCTAOnOffObservation objects can be simply appended to the GObservations container (see #1669). This means that GCTAOnOffObservation now derives from GObservation.

• there is no complete unit test that checks the On/Off fitting functionality. Such a test should be added to the test_CTA.py script (there is already a unit test, but it does so far not perform On/Off fitting)
• the computation of the Hessian in GCTAOnOffObservation::likelihood() looks strange. Could you please write down the intended formula in the documentation (can use LaTeX style), and then check whether the formula is indeed implemented?
• the GCTAOnOffObservation::compute_alpha method actually computes the ratio between effective areas, not exposures. Am I correct that it should compute the ratio of exposures? What about the solid angle?
• in the GCTAOnOffObservation::compute_rmf method, if a weighting is done, should it not be done over effective area times livetime? Where is the time coming in? Could you write down the formula in the documentation what the method should be doing? Also, should the aeff method not use true instead of reconstructed energy?
• shoudn’t the model_on() and model_off() methods be private (it’s always simpler to make a private method public when really needed then vice versa)
• are the alpha() and offtime() methods needed?

Furthermore I see that the PHA data format requires an EXPOSURE keyword (the integration time (in seconds) for the PHA data, assumed to be corrected for deadtime, data drop-outs etc.). So this is a kind of livetime. I think we should add this attribute to the GPha class, and then the storage of ontime and offtime is not longer needed. By the way I see

m_ontime  = obs.livetime();
m_offtime = obs.livetime();

Merge the branch now in the trunk to prevent further divergence, but there are still pending issues that should be resolved as soon as possible (see #1781).

Update of the test files to work with the latest version of the software.