[
https://issues.apache.org/jira/browse/FLINK-1979?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14530471#comment-14530471
]
Till Rohrmann commented on FLINK-1979:
--------------------------------------
Hi [~jguenther], great to hear that you wanna picks this topic up. We're
currently developing an optimization framework which can tremendously benefit
from more loss functions. See FLINK-1889 and FLINK-1807 for more details.
Here is [https://github.com/apache/flink/pull/613] the corresponding pull
request with the current state. We hope to merge it in the next days. You find
the interface for the loss functions in the file LossFunction.scala. There are
also an implementation for the squared loss function.
Be aware that the prediction function will still change a little bit. But they
will more or less have the following interface
{code:Scala}
trait PredictionFunction {
def predict(x: Vector, weights: WeightVector): Double
def gradient(x: Vector, weights: WeightVector): Vector
}
{code}
So in order to implement new loss functions you simply have to implement the
abstract methods.
Shall I assign this issue to you, Johannes?
> Implement Loss Functions
> ------------------------
>
> Key: FLINK-1979
> URL: https://issues.apache.org/jira/browse/FLINK-1979
> Project: Flink
> Issue Type: Improvement
> Components: Machine Learning Library
> Reporter: Johannes Günther
> Priority: Minor
> Labels: ML
>
> For convex optimization problems, optimizer methods like SGD rely on a
> pluggable implementation of a loss function and its first derivative.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
