Negative values for births and deaths

[christian-oreilly]

I apologize in advance if this is explained somewhere or if it is a trivial question. I admit being very new to topological analyses. Anyway, I could not find anything (e.g., some normalization or transformation) that would explain the fact that I obtain negative values for births and deaths on the persistence plot I obtained with Ripser. This happens using the following code:

from ripser import Rips
rips = Rips(maxdim=4)
diagrams = rips.fit_transform(data, distance_matrix=True)
rips.plot(diagrams)

which produced the following figure: https://pasteboard.co/IGNZumH.png

Let me know if this is a bug, in which case I'll share the data that are used in this example so that you can reproduce the issue. If it is not a bug, maybe the documentation can highlight more clearly the fact that the plots generated are not standard persistence plots (i.e., with positive births and deaths) and in which way they depart from standard persistence plots.

[sauln]

oh very weird. Thanks for bringing this up.

Does your distance matrix have negative values?

[christian-oreilly]

Right... I was using a correlation matrix, so you are totally right... with both positives and negative values. My bad. If it is expecting a distance metric it is not surprising that it would give weird results. Maybe I could suggest testing for positivity and throwing a warning if the input matrix has negative values... for dumb-proofing the function call and helping spotting eventual issues :) Thanks for the fast feedback and for your time.

[sauln]

Phew! I'm glad that was the problem.

The test and warning is a great idea and would probably save others in the future. If you'd like to submit a pull request with that implementation, I would readily review and merge!

[ctralie]

Note that our lower star image filtration often applies to negative scalar images, so that would trip the warning erroneously every time. It could be messy to deal with that

…

On Sat, Nov 16, 2019, 1:24 AM Nathaniel Saul ***@***.***> wrote: Phew! I'm glad that was the problem. The test and warning is a great idea and would probably save others in the future. If you'd like to submit a pull request with that implementation, I would readily review and merge! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#88?email_source=notifications&email_token=AAJWDZVCXJBWRAVRCUWOUOTQT6GX7A5CNFSM4JN24V52YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEEHKWTQ#issuecomment-554609486>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJWDZVY7L3PWJFYE3HUSFLQT6GX7ANCNFSM4JN24V5Q> .

[christian-oreilly]

@sauln I can have a look at it.

@ctralie Could you provide an example of a function call that relates to that? Not knowing that much about ripser, it could take me some time to figure out what objects/functions/methods are related to your "lower start image filtration".

[ctralie]

We have two notebooks showing this use case https://github.com/scikit-tda/ripser.py/blob/master/docs/notebooks/Lower%20Star%20Image%20Filtrations.ipynb https://github.com/scikit-tda/ripser.py/blob/master/docs/notebooks/Lower%20Star%20Time%20Series.ipynb Because of this, I would vote against putting a warning for negative values, but thank you for offering!

…

On Sat, Nov 16, 2019 at 8:53 AM christian-oreilly ***@***.***> wrote: @sauln <https://github.com/sauln> I can have a look at it. @ctralie <https://github.com/ctralie> Could you provide an example of a function call that relates to that? Not knowing that much about ripser, it could take me some time to figure out what objects/functions/methods are related to your "lower start image filtration". — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#88?email_source=notifications&email_token=AAJWDZQJGH7PTHMVXAMCRA3QT73OXA5CNFSM4JN24V52YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEEHSA2I#issuecomment-554639465>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJWDZS6KK4W7TOLU7DSQ7LQT73OXANCNFSM4JN24V5Q> .

[christian-oreilly]

Alright. Nice notebook @ctralie

So, no change for the moment. At least this ticket provides a minimum of documentation on the existence of negative birth and death time.