Fix bad thumbnails

[bhilbert4]

This PR makes some tweaks to the preview image generator in an effort to finally fix the bad thumbnail images that have been plaguing us forever.

Previously, the thumbnail images were created using matplotlib's imshow() function. This function creates an AxesImage object. However, preview images are generated by first creating a matplotlib Figure object, and then populating the axes of that object with an AxesImage object via imshow(). It turns out that plt.close only closes Figure objects. This means that thumbnail images were not being closed at the end of the preview image generation process, and this must have been leading to multiple images open at the same time, resulting in bad thumbnail images.

This PR changes the thumbnail creation to be like that of the preview images. We first create a Figure object and then populate that using imshow. Local testing on several programs showed no bad thumbnails while using this. We should try a larger scale test on the server, but it looks promising that this PR will put an end to our bad thumbnail images.

I also fixed a small bug that was preventing some thumbnails from being created, where NaN values in the data were causing bad min/max values for the display.

Resolves #958

[pep8speaks]

Hello @bhilbert4, Thank you for updating !

• In the file jwql/utils/preview_image.py:

Line 50:1: E402 module level import not at top of file
Line 51:1: E402 module level import not at top of file
Line 52:1: E402 module level import not at top of file
Line 453:5: E303 too many blank lines (2)

Comment last updated at 2022-07-22 17:29:15 UTC

[bhilbert4]

We also need to explicitly set the figure size for thumbnail images now, with these changes. At the moment that is hardwired to be 3x3. Is this appropriate? We don't want the figures to be much larger than necessary, since that would slow down image loading and therefore the JWQL webpages.

[bhilbert4]

Tests on the test server were successful. I didn't see any bad thumbnails when using this code. Also, running generate_preview_images is a lot faster when it ignores all of the guide star tracking files.

I feel like this has been a long battle that @BradleySappington, @mfixstsci, and I have been fighting. I'm excited for it to be over. 🤞

@mfixstsci I think this is ready for review.

[BradleySappington]

I'm absurdly excited

[mfixstsci]

@bhilbert4 @bhilbert4 lets go! I am going to start my review

[mfixstsci]

Looks good @bhilbert4 I know you and @BradleySappington have been working hard on this and hope this solves it once and for all.

[bhilbert4]

Fixed a small conflict that came about after the merge of #968