Feature/general response

[avalluvan]

Preview
DetectorResponse class has been expanded with new functionalities
FullDetectorResponse class has been cleaned up, comments have been added, and interpolation of NuLambda axis has been made possible.

DetectorResponse._set_mapping

• Sets the mapping between target value (physical units) and axes labels (any binning scheme that is employed) to facilitate interpolated response calculation
• Generalized implementation to allow testing of various response file binning schemes

_get_all_interp_weights

• Get weights for interpolation along multiple axes for target value provided as a dictionary target
• Returns in the same format as HealpixAxis and Axis. (pixels, weights)

transform_eps_to_Em and transform_Em_to_eps

• Placeholder function to transform between eps coordinates (new binning scheme) and Em coordinates (physical units)
• Generalized implementation to allow expanding to other transformations
• Can also be implemented for other axes

get_interp_response

• Calculates interpolated response value for a target value provided as a dictionary target
• Example: target = {'Ei': 511*u.keV, 'Em': 600*u.keV, 'Phi': 12*u.deg, 'PsiChi': 385}
• Can calculate vector along 'Ei' axis if only three target values are given.
• Currently does not support vector returns along other axes. As the other axes are part of the measured data space, do we really expect to require that?

FullDetectorResponse.getitem

• Restructured code to do away with repeated logic

self.unbinned

• Just a placeholder variable for now. It can be used to impose restrictions on which features are accessible for binned vs unbinned analysis

get_point_source

• Added interpolation code along NuLambda axis that was on the TODO list
• deepcopy statement should be removable but I do not know how to do it. The coordsys fields are in conflict dr.axes['PsiChi'].coordsys == response[p].axes['PsiChi'].coordsys returns False.

Suggested changes for histpy

• Interpolations in log-space should not follow x-x1, x2-x convention. We should switch to x/x1, x2/x convention.
• FullDetectorResponse AXES/TYPE attribute is inconsistent
  • For healpix, it is “healpix”. For axis, it is “linear” or “log”.

Questions about next steps

• Do we expect to reparametrize the model space?
• Should reparametrization handling codes be included in the same DetectorResponse class or should we branch out to a new file? Right now, DetectorResponse is fully back-compatible
• How do you vectorize eps_to_Em conversion when the function parameters change for different values of Ei
• Should we try out preferential weighting along diagonal direction (and reduce the weights along anti-diagonal direction)
• Do we want to use sparse matrices for scatt_map?
• How do we remove the deepcopy statement in FullDetectorResponse.get_point_source()
• When will we deprecate exposure_map in favor of coords + scatt map for both local and inertial coords?

[israelmcmc]

@avalluvan I'll come back to this PR after #369. I feel that a lot of the goals of this PR can benefit from the interfaces refactoring.