Not finding correct elbow about kneed

[MattConflitti]

The array is about 6000+ elements. If I have some time, I will do that. Thanks for the confirmation!

[MattConflitti]

Thanks for looking into this issue and fixing it! Kneed is the first package I have seen to be able to reliably find an elbow or knee programmatically. Very helpful.

…

On Sat, Jul 6, 2019, 7:06 AM Kevin Arvai ***@***.***> wrote: Closed #31 <#31>. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#31?email_source=notifications&email_token=ADBVIG2W2332AUWA3CLNNCLP6B4CRA5CNFSM4H35SFCKYY3PNVWWK3TUL52HS4DFWZEXG43VMVCXMZLOORHG65DJMZUWGYLUNFXW5KTDN5WW2ZLOORPWSZGOSLOAQTI#event-2463893581>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADBVIG5675NPSSG5WBLG7W3P6B4CRANCNFSM4H35SFCA> .

[arvkevi]

Hi @MattConflitti thanks for posting, would you mind sharing which version you are using?

import kneed
print(kneed.__version__)

^^^somehow I managed to not include this in the source code^^^
I'll make a PR to update this.

You could try this from the command line though:
pip freeze | grep kneed

[MattConflitti]

It looks like that attribute wasn't set.

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-63-b9a19a10a516> in <module>()
      1 import kneed
----> 2 print(kneed.__version__)

AttributeError: module 'kneed' has no attribute '__version__'

I installed with pip and that says I have kneed==0.3.1

[arvkevi]

Great, that's the latest version. The last question, have you tried setting interp_method='polynomial' in KneeLocator?

[MattConflitti]

I did, this is the result. Similar to the previous one.

[arvkevi]

Thanks for checking.
Do you want to post x, y here so we're working off the same arrays?

You are correct, the knee should be located at the max of the distance curve yd.
Feel free to hack away on this bug with a pull request, otherwise, I'll try to reproduce the errors as soon as a can.

[arvkevi]

It should be easy enough to reproduce, thanks again, hope to jump on this soon.

[arvkevi]

This x and y reproduce a similar error:

n = 6000
x = range(1, n+1)
y = sorted(np.random.gamma(0.5, 1.0, n), reverse=True)

[arvkevi]

This sparked a long-overdue refactor, working on this in #34

[arvkevi]

Hi, @MattConflitti thanks again for raising this issue, the package really needed this update! I spent some time re-reading the paper to ensure that I understood the algorithm correctly. The issue was that kneed was finding all knees and reporting the most recent (or last) knee point. As a result, setting the S parameter had no effect because the last knee was always returned.

I added your example to the README in #34 because it highlights how tuning the S parameter has an effect on the detected knee.

[arvkevi]

@MattConflitti going to close this for now, but I'm happy to re-open if this doesn't solve the issue.