pyLikelihood Unbinned and Binned Analyses

Notice: The Fermi code has not been released to the public and is for use by members of the LAT Collaboration only!

The pyLikelihood package enables you to write a python script that, in its simplest form, reads in a source model and the event and spacecraft data files, then determines the goodness of fit, i.e., how well the model fits the data. The package is an interface to underlying code written in C or C++ that enables you to perform either an unbinned or a binned analysis by running gtlike, thereby making it an interface to the gtlike tool as well.

The UnbinnedAnalysis.py and BinnedAnalysis.py modules each contain two classes: UnbinnedObs and UnbinnedAnalysis in the former, and BinnedObs and BinnedAnalysis in the latter. Since the respective analyses require different sets of input data, the "Obs" classes differ in how they are constructed.

Notes:

• To better understand the terms and concepts used in this Fermi LAT pyLikelihood tutorial, it is highly recommended that you first perform the Likelihood: Unbinned and the Likelihood: Binned tutorials.

• Before beginning your analysis, refer to:
  Standards for High-Level Analysis.

Comparison of Unbinned and Binned
Analysis Classes and Objects

The UnbinnedAnalysis and BinnedAnalysis constructors are identical except that the BinnedAnalysis constructor takes an instance of BinnedObs as the observation object.

Factoring into separate analysis and observation classes affords flexibility in mixing and matching observations and models in a single Python session or script while preserving computational resources, so that one can do something like this:

>>> analysis1 = UnbinnedAnalysis(unbinnedObs, "model1.xml")  
>>> analysis2 = UnbinnedAnalysis(unbinnedObs, "model2.xml")

Even though the two UnbinnedAnalysis instances access the same data in memory, distinct source models may be fit concurrently to those data without interfering with one another. This is useful in comparing, via a likelihood ratio test, for example, how well one model compares to another.

Analysis Classes

Objects