Github user MLnick commented on the issue:
https://github.com/apache/spark/pull/11974
@sethah the idea is to get to a "good enough" solution in a fraction of the
time. The [paper](http://www.eecs.tufts.edu/~dsculley/papers/fastkmeans.pdf)
shows it achieving "near optimal" cost in far fewer iterations. Scikit-learn
has
[MiniBatchKMeans](http://scikit-learn.org/stable/modules/generated/sklearn.cluster.MiniBatchKMeans.html)
and shows a comparison with KMeans that indicates the resulting clustering is
typically [very
similar](http://scikit-learn.org/stable/auto_examples/cluster/plot_mini_batch_kmeans.html).
Though @zhengruifeng the above comparison should probably also compare the
"fully converged" full data solution (cost and # iters) with the minibatch
solution (cost and # iters). It is not clear from what you posted what that
result is?
Actually I also forgot that we will definitely need to update the examples
and also add a section in the guide explaining the use of `miniBatchFraction`
and explaining the tradeoffs and providing references...
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
