TST: fix some mpl warnings

[mzeitlin11]

• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff

This change fixes UserWarning being emitted in tests calling autocorrelation_plot, andrews_curves, and radviz.

UserWarning was caused by calling plt.gca(xlim=..., ylim=...) on a potentially already existing axis, so matplotlib warns about this forcing new axis creation. Solution was just to explicitly use set_xlim and set_ylim instead.

Does it make sense to also add to the relevant tests an assertion that no UserWarning is emitted to guarantee the new behavior?

[ivanovmg]

Please see the comment below.
I agree that addition of the tests is necessary.

[ivanovmg]

I guess you can even use ax.set_xlim(-1, 1), without extra brackets.
The same for other cases as well.

[mzeitlin11]

Done

[jreback]

yeah pls add a tests that asssert_produces_warning(None) and fails on master but passes here

[mzeitlin11]

Will add tests once #38620 is figured out

[jreback]

Will add tests once #38620 is figured out

hmm tm.assert_produces_warning(None) should just work

[mzeitlin11]

Will add tests once #38620 is figured out

hmm tm.assert_produces_warning(None) should just work

It wasn't raising even when a warning was returned, with examples of this given in the issue. So there are currently no failures in master when wrapping these plotting calls with tm.assert_produces_warning(None), even though a warning is returned

[mzeitlin11]

Wrapped existing calls that give warnings on master to avoid adding more plotting calls. If splitting out new tests for this is preferred instead, please let me know

[jreback]

Wrapped existing calls that give warnings on master to avoid adding more plotting calls. If splitting out new tests for this is preferred instead, please let me know

this looks good; ping on green

[mzeitlin11]

Green

[jreback]

thanks @mzeitlin11

[simonjayhawkins]

@meeseeksdev backport 1.2.x