update selection of interpolation methods

[jimustafa]

Attempts to address #28.

• replace 'none' with blackman

Something else to consider would be making the image data the same as what is used in the current matplotlib gallery example for the interpolation methods. To that effect, I could add a commit to this PR.

[jklymak]

I actually am not sure we should advertise these so exhaustively. It takes a lot of space for something that is pretty advanced.

'none' otoh probably should not be dropped?

[jimustafa]

It seemed that there was some consensus in #28 that 'none' could be dropped. For clarity, could we list the 18 interpolations that should be demonstrated.

While I agree that a 3x3 grid of 9 common interpolations may be sufficient, that does leave the question of what to put in the resulting open space. Maybe this is best addressed in a separate issue/PR (see #81).

[jklymak]

I suspect the right thing to do here is to make sure the pcolormesh objects are rasterized so that the result of "none" does not depend on the pdf viewer. However "none" is very heavily used (for better or worse) so I think it would be odd to not document how it works in the cheat sheet.

[jklymak]

@jimustafa are we publishing the artifacts from CI anywhere?

[jimustafa]

Yes @jklymak, the cheatsheets and handouts are saved as build artifacts during workflow execution, and published to the gh-pages branch after being successfully built.

cheatsheets/.github/workflows/main.yaml

Lines 57 to 70 in efd4431

	- uses: actions/upload-artifact@v2
	if: ${{ always() }}
	with:
	name: build
	path: |
	cheatsheets.pdf
	handout-*.pdf
	- name: Publish cheatsheets and handouts
	if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
	uses: peaceiris/actions-gh-pages@v3
	with:
	github_token: ${{ secrets.GITHUB_TOKEN }}
	publish_dir: ./build/
	force_orphan: true

[jklymak]

but only if its a merge to master, right?

[jimustafa]

While the artifacts are saved for each build, publishing to gh-pages is only on merge to master, yes.

[jklymak]

Yeah, but I never know where the artifacts are stored. Probably just me not seeing a button somewhere...

[QuLogic]

They're on the Summary page, e.g., https://github.com/matplotlib/cheatsheets/actions/runs/1433495962

[jimustafa]

Hey @jklymak. Is this what you were suggesting?

[jklymak]

I think so. Did it work? After suggesting it, I wasn't sure if imshow rasterized, but I'm pretty sure it does.

[jklymak]

This looks fine in my viewer. I think rasterizing it is the right way to go to make it clear what the differences are.

OTOH, the reader is definitely going to be confused as to what None/none/nearest all mean in this context. If I were doing this I would have a section:

# Image interpolation (over-sampled)

"nearest"/"none" = default ("antialiased", `None`), "bilinear"; "bicubic"; etc..

and a different section

# Image interpolation (under-sampled)

## `interpolation_stage="data"`:

"antialiased" = default(`None`); "none"; "hanning"; etc (the other low-pass filters don't matter a ton)

## `interpolation_stage="rgba"`:

"antialiased" = default(`None`); "none"; "hanning"; etc (the other low-pass filters don't matter a ton)

the fact of the matter is that for over-sampled images (many more pixels than data) "none" is a perfectly sensible choice. For under-sampled data (fewer pixels than data) "none" is almost always aliased and should be low-passed first.

See https://matplotlib.org/devdocs/gallery/images_contours_and_fields/image_antialiasing.html

[timhoffm]

I think this is the right fix for a consistent display.