[ENH] Wrapper for hmmlearn

[miraep8]

Related to my issue #3099 on how I plan to further improve our support of HMM-like algorithms in sktime and also the issue I opened in hmmlearn (#484)

Will make a non-draft PR when it is ready for review. At a high level the plan is o have a simple base class that interfaces with the BaseAnnotator class and provides interface compatibility. Then each HMM-like class which inherits from this base class simply imports the relevant HMM estimator from hmmlearn.

Still to do:

• make sure parameter setting is consistent/works!
• add some more tests that things work as expected.

[fkiraly]

All looks good, except the implicit change to the interface contract in "object".

Is there a way without changing the heart of sktime interface specs?

Worst case, lots of typing of self.=...
for a slightly less exhausting route, have a look at the ARIMA class.

[miraep8]

Removed the attrs bit now and still seems to work ok

[fkiraly]

Looks good, but your doctest is failing.

If you want expressions that continue over multiple lines, you need to use ... from the 2nd line on, instead of >>>

[KatieBuc]

Looks great @miraep8!

[fkiraly]

Code looks good to me!

May I ask for some changes re soft dependency handling, the general rule is "don't import soft dependencies directly, anywhere". So:

• in the test, use a pytest.skipif and import hmmlearn inside the test. Compare how pmdarima is handled, search for pytest.skipif.
• set the python_dependencies tag in the new classes (base class is enough)
• skip the example doctest (search for # doctest to see examples). If this reduces coverage, move the examples into another test decorated with pytest.skipif.

[miraep8]

Since GMMHMM has already been merged and that PR was a superset of this one I am going to close this PR.

[fkiraly]

I see, I should have merged them in a sequence.

Sorry for this.

Next time, may I kindly ask to write sth explicit like "contains PR # 3156" in the other PR instead of sth flowery like "building on success in #3156" (which can simply mean you are following the same pattern).