Issue 385 CLI: show beliefs does not handle NaN well

[Flix6x]

Closes #385.

[coveralls]

Pull Request Test Coverage Report for Build 3350902973

• 6 of 20 (30.0%) changed or added relevant lines in 3 files are covered.
• 2 unchanged lines in 1 file lost coverage.
• Overall coverage decreased (-0.07%) to 64.949%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
flexmeasures/utils/time_utils.py	2	4	50.0%
flexmeasures/data/models/generic_assets.py	1	4	25.0%
flexmeasures/cli/data_show.py	3	12	25.0%

Files with Coverage Reduction	New Missed Lines	%
flexmeasures/init.py	2	81.82%

Totals
Change from base Build 3259160910:	-0.07%
Covered Lines:	6514
Relevant Lines:	9413

---

💛 - Coveralls

[nhoening]

Great. Two comments you can consider.

[nhoening]

In my PR in flexmeasures-simulate, I have a similar util function, and I decided to let it raise if the resolutions are actually not compatible (if any([resolution % min_resolution != timedelta(hours=0) for resolution in resolutions])

This function could do this, maybe triggered by a parameter flag raise_if_incompatible. Thoughts?

[Flix6x]

In my mind, no combination of resolutions should be incompatible. Resampling a (simplified) time series with instantaneous values should be rather straightforward: when downsampling, select the first value; when upsampling, interpolate using some strategy (such as forward fill or linear interpolation), which we could later define as a sensor attribute. I'll address this soon in a timely-beliefs PR.

[nhoening]

I couldn't see how you use uniplot differently now. It would probably still work with <0.7.0, just not if NaN is in the data. So, do we technically need to depend on a higher version, or do we like this feature to be around? Always good to add the reasoning as a comment to make this clear for later.

[Flix6x]

I would like this feature to be around. When selecting multiple sensors for some time period, having NaN data in that set is very common. With previous uniplot versions, the time series will not align right, which is a serious problem to me.

From uniplot's setup.py (version 0.7.0):

python_requires=">=3.5",
install_requires=["numpy>=1.15.0"],

which are older versions than the first ones we started using for this repo as far back as I can tell (10cff9f).

I'll add an inline note to where we define the minimum version requirement.