Confused by DBSCAN min points

[uhho]

Confused by DBSCAN min points

[uhho]

@awenger Thanks for your comment!

As far as I understand the algorithm, this is correct behaviour.
According to algorithm description, minPts is a minimum number of points in ε-neighbourhood around the core point to form a cluster.
In your case, ε is was too small to form a cluster containing at least 3 points, so all points were marked as a noise.

Yes, I agree with you it may be confusing and all parameters needs brief explanation.
I've got two ideas:

• write more detailed explaination it in JSDoc comments included in the code
• add API.md containing list of exposed functions and detailed explaination

What do you think?
If you want to help, feel free to submit pull request!

[awenger]

@lukaszkrawczyk I'm also not 100% sure. But if I look at wikipedia it describes the parameters as: DBSCAN requires two parameters: ε (eps) and the minimum number of points required to form a dense region [1]

The pseudocode for regionQuery also includes the point that was used to query the neighborhood [1]:

regionQuery(P, eps)
   return all points within P's eps-neighborhood (including P)

I guess this is different in this DBSCAN implementation: https://github.com/LukaszKrawczyk/clustering/blob/master/lib/DBSCAN.js#L184

What do you think?
[1] http://en.wikipedia.org/wiki/DBSCAN

[uhho]

Yes... there was a similar discussion on including core point in OPTICS algorithm some time ago:
#4.

I checked how it's implemented in other libraries (such as e.g. SciKit), and I'm more and more confident we should change the algorithm to include core points, both in DBSCAN and OPTICS.

I will modify code and update package on NPM.

[awenger]

Thanks a lot for the update