docs: PCA and UMAP OutlierDetection docstring examples

[anopsy]

Docs

Added usage examples to:
decomposition.umap_reconstruction.UMAPOutlierDetection
decomposition.pca_reconstruction.PCAOutlierDetection

Fixes #652 and #653

Type of change

• Bug fix (non-breaking change which fixes an issue)
• [ x] New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)

[anopsy]

Wondering what happend here

[FBruzzesi]

Just wondering, how did you come up with these array values?
Wouldn't make sense to have some randomly generated arrays (with fixed seed) and then tinker around some value to make them outliers? Point is: currently I cannot spot outliers just by looking at the arrays and everything seems a bit magical.

On the Github action failing: Getting a sense that macos-latest runner is on macos 14, and it is breaking most of github actions.
I can't deep dive today, but will try in the coming week.

Anyway, certainly unrelated from the docstring changes

[anopsy]

Oh gee, there are even more failed tests now.😮

About examples
I copied the approach with the arrays from scikit-learn docs on PCA (check ss)

and I tinkered around with the values to check how the detectors work for different values in the arrays, n_components and thresholds. I tried to add values that would be clearly an outlier because of the quantiles for example [-100, 99, -99] but the PCA/UMAPOutlierDetectors "classified" them as inliers. I also saw in User Guide that the values classified by PCA/UMAPOutlierDetectors as outliers, don't look like quantiles based outliers -so you can't spot them just by looking. If that makes any sense.

[FBruzzesi]

Yep the doc page is using iris dataset, which I would not expect to have any particular outlier.
We have one obvious example in the test suite.

@koaning thoughts on this? In my opinion, it could be worth it to change dataset in the user guide as well. It seems a bit confusing

[anopsy]

Sure, I'll do it the way it's done in the test suite.

[anopsy]

Yep the doc page is using iris dataset, which I would not expect to have any particular outlier. We have one obvious example in the test suite.

Hey Francesco I'm back at it. I was having some thoughts about the examples for PCA and UMAPs, what bugs me is that if I use the obvious example which is a 10-d array, how can I show the resulting outliers? Should I print the 10d output?
I mean that the numbers in the arrays I used may seem arbitrary, but at least we can show the outliers in a simple one line output.
Let me know wdyt

[FBruzzesi]

Let's ship it ✨