0
$\begingroup$

Mahout, Hadoop machine learning library, contains an implementation of Streaming K-means algorithm that is based on the following paperworks The Effectiveness of Lloyd-Type Methods for the k-Means Problem and Fast and Accurate k-llleans For Large Datasets.

By analyzing this approach I came up to an idea to implement K-medoids algorithm in a similar fashion like Streaming K-means.

I'm looking for a validation of this idea - can someone give me opinion about it, is it a good idea, and if not, what are the flaws?

Improve this question
$\endgroup$
3
  • $\begingroup$ k-means in Mahout is so incredibly slow. Dump Mahout, don't waste your time on it. It's just not meant to be used with mapreduce. $\endgroup$
    – Has QUIT--Anony-Mousse
    Commented Jun 1, 2015 at 20:30
  • $\begingroup$ @Anony-Mousse Thanks for the tip. What do you suggest instead? By the way, I'm thinking about implementing a similar algorithm myself, not using Mahout... $\endgroup$
    – Kobe-Wan Kenobi
    Commented Jun 1, 2015 at 20:55
  • $\begingroup$ Run you own benchmarks. Try different tools, to figure out what works for you and is competitive. Then build on top of that. I'd be interested to see how Mahout fares in your benchmarks. $\endgroup$
    – Has QUIT--Anony-Mousse
    Commented Jun 1, 2015 at 21:20

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.