[WIP] Setting weight coefficients in OPTICS algorithm.

[trabbart]

Reference Issues/PRs

See issue #12394

What does this implement/fix? Explain your changes.

This implement settings weight coefficients in OPTICS algorithm. Changes are similar to solution in DBSCAN.

Any other comments?

[adrinjalali]

Thanks for your PR @trabbart

there are a few issues which need to be addressed here:

• sample_weight should be passed to fit, and not init
• you need to add tests for the functionality you add
• I'm not sure how to handle min_samples when we have sample_weight.
  Should min_samples be the weighted number of samples?

[trabbart]

Thanks for your review @adrinjalali

I will add test and correct mistake about passing sample_weight.
About third dot I think is good to make it consistent with implementation in DBSCAN.
In DBSCAN min_samples is required minimal sum of weight of samples in cluster (Is this what you call “weighted number of samples”?).
There is one difference between OPTICS and DBSCAN – in OPTICS min_samples can also be float. (Then it is interpreted as fraction of number of all samples). I propose to treat in similar way. So minimal sum of weight will be equal to fraction of sum of weights of all samples. I am not sure if this is intuitive. What do you think?

[adrinjalali]

There is one difference between OPTICS and DBSCAN – in OPTICS min_samples can also be float. (Then it is interpreted as fraction of number of all samples). I propose to treat in similar way. So minimal sum of weight will be equal to fraction of sum of weights of all samples. I am not sure if this is intuitive. What do you think?

yeah that makes sense

[armenabnousi]

Hi,
This issue is closed but it's not on the dev version yet. Would appreciate if you could give a timeline of when it will be integrated.

Thanks

[adrinjalali]

I don't think anybody's actively working on it, so there's no ETA at the moment.

[MitalPattani]

@adrinjalali @trabbart has used radiusNeighbors instead of simply kneighbors for sample weights what is the purpose behind that? and won't that be inconsistent?