Initial tests of FDNT::RunFDNT() in MegaLUT framework

[reikonakajima]

Similar for the case of GLMoment(), but for the full, PSF-corrected RunFDNT() instead.

[reikonakajima]

fdntfunc.py was moved under megalut/meas where it belongs.

[reikonakajima]

The general framework is now working (commited fdntfunc.py in megalut/megalut#68).

However, there are several remaining issues before shape measurement accuracy can be tested.
For the brightest galaxies (flux = 20000! normally ~100) in 900 simulated galaxies, I get the following flags:

• 1 (Edge), 67 counts
• 9 (CenteroidMismatch + Edge), 344 counts
• 1025 (OutOfBounds + Edge), 361 counts
• 8320 (GLFailure + TooSmallTrustRad), 91 counts
• 8321 (GLFailure + TooSmallTrustRad + Edge), 37 counts

I need to investigate where they are coming from, and generate a workaround.

[reikonakajima]

Edge:
From FDNT, it originates in prepare(), based on flags of exposure groups, in ExposureGroup::prepare().
From GLSimple2, it originates in acquireData().

CentroidMismatch:
From FDNT, in 3 different places: marginalizedCentroid(), processSample(), and shape2() (when integrating over all probabilities, if this probability sum is larger than CENTROID_FLAG_THRESHOLD).

OutOfBounds:
From FDNT, it comes from ExposureGroup::prepare() (5 different cases, has to do with FFT size requirements)
From GLSimple2, comes from acquireData(), if image not found within bounds

GLFailure:
From FDNT, it comes from GLAll() when GL fitting fails on the native basis fitting on all exposures of the single galaxy.

TooSmallTrustRad:
From GLSimple3::solve(), when fitting algorithm fails at a specific point in setting the trust radius.

[reikonakajima]

Tried larger galaxy and PSF, so that PSF fit would succeed. However, TooSmallTrustRad is always triggered upon PSF measurement.

[reikonakajima]

First problem solved: The PSF image was not centered properly (because I was drawing it onto a FITS file, which requires the origin (bottom left corner) to have (1,1) coordinates).

I've removed the code that tinkers with the TrustRadius and returned it to its original state. Now there are only two types of errors that come out with the "normal" FDNT code, with a large PSF:
Either (1) fails with flag 1025 (OutOfBounds+Edge), or (2) succeeds with flag 1 (Edge).

[reikonakajima]

All of the OutOfBounds error seem to stem from the following error:

 hypot < MINIMUM_FFT_SIGMA in x (or y)

It seems I will need to pad the images so that this FFT_SIGMA is met.