[ENH] Adding min_periods in WindowSummarizer to avoid different behaviour in the future

[arnavrneo]

Reference Issues/PRs

Fixes #4050

What does this implement/fix? Explain your changes.

In the WindowSummarizer class, min_periods is explicitly defined in the pandas rolling function call with the value equal to the window_length.

Does your contribution introduce a new dependency? If yes, which one?

No.

[arnavrneo]

@KishManani @miraep8 Kindly look into and review this PR.

[aiwalter]

@arnavrneo do you mind to add a test for this?

[fkiraly]

• linting is failing, see the guide here on how to format your code: https://www.sktime.org/en/stable/developer_guide/coding_standards.html
• not sure whether I understand the deprec message correctly, maybe I am confused - should the min_periods not be set to 0 rathre than window_length?

[KishManani]

not sure whether I understand the deprec message correctly, maybe I am confused - should the min_periods not be set to 0 rathre than window_length?

On pandas v1.5.2 the default behaviour for something like df.rolling(window=3).mean() is to do df.rolling(window=3, min_periods=3).mean(). Changing to df.rolling(window=3, min_periods=0).mean() would give you a different behaviour.

import pandas as pd
import numpy as np

y = pd.DataFrame(data=np.random.rand(10))
y.rolling(window=3).mean()
	0
0	
1	
2	0.7207749584147084
3	0.5946349404144744
4	0.34069940037566454
5	0.30547069208136307
6	0.5604781957810597
7	0.7837284354272988
8	0.6594005203859353
9	0.44447740298324145

y.rolling(window=3, min_periods=3).mean()
	0
0	
1	
2	0.7207749584147084
3	0.5946349404144744
4	0.34069940037566454
5	0.30547069208136307
6	0.5604781957810597
7	0.7837284354272988
8	0.6594005203859353
9	0.44447740298324145

y.rolling(window=3, min_periods=0).mean()
	0
0	0.46538184566621255
1	0.7205428982372711
2	0.7207749584147084
3	0.5946349404144744
4	0.34069940037566454
5	0.30547069208136307
6	0.5604781957810597
7	0.7837284354272988
8	0.6594005203859353
9	0.44447740298324145

[arnavrneo]

@aiwalter @fkiraly Okay, I'll again run the tests and will update the PR asap.

[fkiraly]

Thanks, @arnavrneo!

Kindly note that linting is failing, see the guide here on how to format your code: https://www.sktime.org/en/stable/developer_guide/coding_standards.html

[arnavrneo]

@fkiraly On my end, tests have formatted the code. Can you please review it again?

[fkiraly]

still not, your setup might be incomplete?
You can also see it on the CI, if you click here:

Due to the release being soon, I will just go through it myself - might be worth if you check what went wrong with your linting setup though.

[fkiraly]

(I think I fixed the linting issues)

[arnavrneo]

@fkiraly Okay sir, I will reset the linting check setup. Thank you for going through the linting yourself.

[fkiraly]

Linting issues are fixed now, thanks for your contribution!