Basic (boxcar) background subtraction

[kecnry]

This PR implements specreduce.background.Background(image, trace/center, separation, width) support.

The current API supports the following pseudo-code (see the edited notebook for full details):

# simplest case to create a background around a row
bg = Background.two_sided(image, ext_center, separation=bkg_sep, width=bkg_width)

# or by defining and passing the spectrum trace
trace = FlatTrace(image, ext_center)
bg = Background.two_sided(image, trace, separation=bkg_sep, width=bkg_width)

# or to only have the background from above/below
bg = Background.one_sided(image, trace, separation=bkg_sep, width=bkg_width)

# or most generally by passing a list of traces
bg = Background(image, [trace-bkg_sep, trace+bkg_sep], width=bgk_width)

# could pass image-bg to automated trace-detection algorithms to refine further

boxcar = BoxcarExtract()
spectrum = boxcar(image-bg, trace, width=ext_width)

In doing so, this:

• exposes some internal methods from Boxcar extract to re-use here
• adds and uses +/- offset support within traces

[PatrickOgle]

Please add the following options, which I consider to be essential:

1. Option to use median rather than average to compute the background
2. Option for 1-sided background extraction.

[camipacifici]

Very nice!
Question:
If I compute the background like this
bkg = Background.two_sided(data, trace, separation=7, width=5)
and I plot it with
plt.imshow(bkg.bkg_image(data), origin='lower')
I get

If instead I use the median
bkg = Background.two_sided(data, trace, separation=7, width=5, statistic='median')
and I plot it in the same way, I get

What am I missing?

[camipacifici]

Typo in the description of the PR:
bg = Background.two_side(image, trace, separation=bkg_sep, width=bkg_width)
two_side -> two_sided

[kecnry]

@camipacifici

If instead I use the median bkg = Background.two_sided(data, trace, separation=7, width=5, statistic='median')...

this should be fixed now. The slicing was collapsing the axes so returning a single value instead of an array, which oddly didn't cause the rest of the code to complain. Now it returns the same shape for the background_image and it seems reasonable on the example data in the notebook. Thanks for noticing that!

[tepickering]

some tests will need to be added before this can be merged.

[ojustino]

This works without issue for me. I see improved performance as the given trace comes closer to the image's actual trace, which is sensible. The test passes.

I think I missed the conversation on the one and two-sided options, but the implementation looks OK to me. I left a couple of more minor comments that shouldn't prevent a merge but could be helpful for the future.

Based on the commented cell in notebook_sandbox/jwst_boxcar/boxcar_extraction.ipynb, it looks like a merge should wait until #85 is finished so the notebook can be updated first.

[tepickering]

looks good to me for now. thanks for adding the test!

[kecnry]

Rebased on top of #85 and added a call to KosmosTrace (using the subtracted image from a FlatTrace) into the sandbox notebook.

[tepickering]

great! i'll merge once CI completes, if that's ok.

[kecnry]

Yep, go ahead and merge if you're happy. I just quickly address @ojustino's suggestions for inline comments and docstrings (sorry that re-triggered CI to restart).