Forced photometry doesn't take into account 0 values in the noise and background maps

[ajstewart]

Current the forced photometry code doesn't seem to be taking into account the border of 0 values you get from the maps selavy produces (this can be in areas where there is image data present).

This causes divide by zero errors:

(vast-tools-dev) ada:~/ada1/temp/forced_phot_testing$ python forced-test.py
WARNING: FITSFixedWarning: 'datfix' made the change 'Invalid parameter values: MJD-OBS and DATE-OBS are inconsistent.
Set MJD-BEG to 58785.569143 from DATE-BEG.
Set MJD-END to 58785.578705 from DATE-END'. [astropy.wcs.wcs]
Removing %i sources that are outside of the image range 0
/suphys/aste7152/.conda/envs/vast-tools-dev/lib/python3.8/site-packages/astropy/units/quantity.py:477: RuntimeWarning: divide by zero encountered in true_divide
  result = super().__array_ufunc__(function, method, *arrays, **kwargs)
/suphys/aste7152/.conda/envs/vast-tools-dev/lib/python3.8/site-packages/astropy/units/quantity.py:477: RuntimeWarning: invalid value encountered in reduce
  result = super().__array_ufunc__(function, method, *arrays, **kwargs)
/suphys/aste7152/.conda/envs/vast-tools-dev/lib/python3.8/site-packages/astropy/units/quantity.py:477: RuntimeWarning: invalid value encountered in true_divide
  result = super().__array_ufunc__(function, method, *arrays, **kwargs)
/import/ada1/aste7152/software/development/forced_phot/forced_phot/__init__.py:807: RuntimeWarning: divide by zero encountered in true_divide
  out = fitter(g, xx[good], yy[good], d[good], weights=1.0 / n[good])
/import/ada1/aste7152/software/development/forced_phot/forced_phot/__init__.py:809: RuntimeWarning: divide by zero encountered in true_divide
  chisq = np.zeros(len(X0)) + (((d[good] - model[good]) / n[good]) ** 2).sum()

One simple fix is to mask these at the data load stage with NaN's so they are caught later on in the NaN check, what do you think @dlakaplan?

        self.bgdata = self.fb[0].data.squeeze()
        self.noisedata = self.fn[0].data.squeeze()
        self.bgdata[self.bgdata == 0] = np.nan
        self.noisedata[self.noisedata == 0] = np.nan

[dlakaplan]

I hadn't encountered this. I think filling with nan is good. I can do that later today.

[dlakaplan]

However, the data may be entirely valid even if the bg map is 0. is it better to set the values to NaN or to set them to e.g., 0 for the bg and the median noise value for the noise?

[dlakaplan]

OK, I think this is pushed to my version. And the try/except for #298 as well. But I don't have the files to test so please do that and let me know

[ajstewart]

Thanks @dlakaplan, these fixes solve the problems.