[ENH], [BUG] utils/plotting.py: change plot_series() to use matplotlib conventions; enhance plot_interval()

[ericjb]

1. sktime.utils.plot_series() now handles xticks following matplotlib defaults. (Fixes [BUG] Specification of xticks for matplotlib axis returned by plot_series does not work as expected #5895) As a result fig,ax objects returned by plot_series() can be used correctly by matplotlib. See accompanying pdf for details including some before and after plot comparisons. (section 3 in the pdf)

2. Fixed (or enhanced) sktime.utils.plot_series() so that it can plot multiple coverage levels if present in the pred_interval multiindex. (section 5 in the pdf)

PullRequestNotes.pdf

[fkiraly]

Thanks!

Docs attached make it clear that it fixes the bug issue.

I am wondering whether and how we could test this - I do not have a good idea though.

[fkiraly]

PS: thanks a lot for the explainer doc! Must have been a lot of work!

I have some comments about the data sets and the issues. I will add more extensive comments in an issue or PR which I expect will open soon? Briefly, major comments:

• categorical variables should be allowed, we are working on extending the framework - see [ENH] umbrella issue - categorical feature support #6109. Also, FYI @Abhay-Lejith, @yarnabrina
• since you are adding the data, it would be great if you could add the metadata with the datasets - you made the table after all, so it would be nice ot make this queriable for users! There is a plan to make datasets queriable similar to estimators with tags etc, so it might be good to have this somewhere. How about dict, {datasetname : tag_dict} with tag_dict = {"univariate": TrueFalse, "scitype": "Series"/"Panel"/"Hierarchical"}? Not required, but if you are doing this very helpful work anyway, I was assuming you have a structure like that floating around in your code? Compare DRAFT [ENH] experimental: datasets as objects #4333, early prototype.