ENH Remember predecessor in OPTICS #12135
Conversation
This will allow filtering the extracted clusters, in order to improve quality (see Schubert and Gertz 2018)
|
Awesome, I think it'd be nice if we had a test to check the |
|
@adrinjalali predecessor, reachability, order (+ a consistency check for core sizes) are tested against ELKI in kno10@abaa35a |
|
Oh yeah, didn't notice the other arrays in that test of yours when I had a look. Cool! Thanks @kno10 |
|
I just added these tests last night, so you probably just saw the first commit of these. |
|
@qinhanmin2014 the test @kno10 to which he's referring IS the output of ELKI if I'm not mistaken. He also has the command with which you can reproduce the results, which is awesome, cause the filter you need to add to process the csv is not trivially explained in the ELKI docs. I think we just need to either create a PR from that, or incorporate it into #12054. |
|
Yes I've seen that. I think we might replace #12054 with that one. |
This will allow filtering the extracted clusters,
in order to improve quality (see Schubert and Gertz, LWDA 2018)
Reference Issues/PRs
Predecessor information was requested in #12049
What does this implement/fix? Explain your changes.
Store the predecessor of each point, which is useful information to filter certain difficult points when extracting clusters (see Schubert and Gertz, LWDA 2018).
Any other comments?
The extraction methods still need to be modified to use this.