Github user zhengruifeng commented on the issue:
https://github.com/apache/spark/pull/16654
@srowen I agree that metric should be irrelevant to details of the
algorithms. AUC is irrelevant to algorithms, it is just relevant to the
dataset: In spark-ml, scikit-learn, or any other packages, the input dataset
contains `label,decision values(or probabilities)`ï¼ if and only if there
exist two labels in the dataset, AUC can be computed, no matter which
classifier is used.
I also agree that some general metrics should be abstracted in Evaluator.
I just disagree that if we treat WSSSE as a general metric:
There have been some attempts to add K-Medoids in spark, although their PRs
were not accepted, there are still some third-party source implementing
K-Medoids on spark.
More realisticly, Spark is used together with other ml-packages in many
cases, suppose use other packages to generate the model locally, and evaluate
the result in spark.
---
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
