Best Method for PairwiseDistance with jagged PDs?

[jcoll3]

I am running KNN classifiers over persistence diagrams using PairwiseDistance. Not all of the diagrams have the same number of persistence points, which is currently required.

As a work around I am currently just padding the diagrams with points at [0,0] to make all have the same number of persistence points, but I feel like that is probably not right. Is there a better way to approach this?

Thanks!

[ulupo]

Hi @jcoll3! Thanks for your presence on this repo and for submitting nice questions/reports.

As you probably know, if the collections of persistence diagrams to be fed to PairwiseDistance were produced in giotto-tda, appropriate padding will have been performed for you. I therefore assume you are unable to produce these collections from within giotto-tda (e.g. because we do not support all filtrations yet).

In that case, I think padding will be the least verbose way to approach this. You should be able to extract what you need from here, please let me know if that's not so clear.

[ulupo]

Of course, the other way is to look at the internals of PairwiseDistance and avoid padding, but this will be substantially harder because PairwiseDistance is so generic.