Implement GLMoments() function Python interface.

[reikonakajima]

Implement a Python interface method, where the native shape can be measured using GL fits.

This should also fix the bogus value return of observedSignificance() in RunFDNT.

[reikonakajima]

Need to update the following files with the new function: fdnt/runfdnt.py (to set defaults) fdnt/__init__.py (to load GLMoments into namespace) pysrc/RunFDNT.cpp (to set Python interface) src/RunFDNT.cpp (to implement the function in C++) src/RunFDNT.h (the header for the two RunFDNT.cpps) examples/demo1.py (a demo script)

[reikonakajima]

The functionality is in there, and it compiles. Now to write some test/demo code.

[reikonakajima]

The weird bug with covE assignment gave the following message

terminate called after throwing an instance of 'img::ImageError'
  what():  Image Error: Destroying ImageData that still has children
Aborted (core dumped)

which is an error message running an Image destructor. It would show up at the most random places --- nowhere where the image was being touched. I would change the code, and it would appear at different places. It seems that it was related to tmv::SymMatrix<double> destruction events, but I'm not entirely sure.

The bug seems to have been fixed with the above commit.

[reikonakajima]

Remaining bugs:

• [ ] e1/e2 variance value is way off (possibly covariance too)
• [x] significance gives NaN

[reikonakajima]

I have a question for @gbernstein (hello!): I'm finally getting back to work on FDNT. I would like to implement the GLSimple::shapeVariance() function, which is declared, but not defined. (note: this is not a FDNT method, but the GLSimple method.) Would you mind jogging your mind a bit to give me a rough outline as to how one would do this? The covariance matrix for the b vector is already there, so perhaps I can use this? Thanks for your help in advance.

[gbernstein]

Hi Reiko - I could probably do more mind-jogging, but I think you're right. The basic idea would be that Cov(e) = D^{-1} Cov(b) D^{-1}^T where D is db/de, the derivative matrix of the b-vectors under application of additional shear. I think it's primarily b20 (real & imaginary) that change under shear so perhaps you would marginalize Cov(b) down to just these two components. The remaining subtlety is that shear doesn't just add to the existing shape e: if the shape is e, and you apply some shear g, the resultant shape e' is not e+g. So there is a de/dg matrix that depends on the original e, and you would need to make sure you have that in the right place depending on what you want "shape variance" to mean.

[reikonakajima]

Hi Gary---Thanks so much! I'll try to proceed from here.