[
https://issues.apache.org/jira/browse/FLINK-1994?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15110605#comment-15110605
]
ASF GitHub Bot commented on FLINK-1994:
---------------------------------------
Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/1397#discussion_r50401844
--- Diff:
flink-staging/flink-ml/src/main/scala/org/apache/flink/ml/regression/MultipleLinearRegression.scala
---
@@ -107,6 +107,11 @@ class MultipleLinearRegression extends
Predictor[MultipleLinearRegression] {
this
}
+ def setOptimizationMethod(optimizationMethod: String): this.type = {
--- End diff --
@chiwanpark, that could also be a solution. Then you could create an
instance of a sub class with the respective parameters (e.g. the decay value).
That makes also more sense, since not all optimization methods depend on the
decay value. I've opened a PR where I did something similar for the abstraction
of the actual calculation scheme (https://github.com/rawkintrevo/flink/pull/1).
This should be easy to extend to include the decay value only for the `Xu` and
`InvScaling` methods. Then we can remove the decay parameter from
`IterativeSolver`.
> Add different gain calculation schemes to SGD
> ---------------------------------------------
>
> Key: FLINK-1994
> URL: https://issues.apache.org/jira/browse/FLINK-1994
> Project: Flink
> Issue Type: Improvement
> Components: Machine Learning Library
> Reporter: Till Rohrmann
> Assignee: Trevor Grant
> Priority: Minor
> Labels: ML, Starter
>
> The current SGD implementation uses as gain for the weight updates the
> formula {{stepsize/sqrt(iterationNumber)}}. It would be good to make the gain
> calculation configurable and to provide different strategies for that. For
> example:
> * stepsize/(1 + iterationNumber)
> * stepsize*(1 + regularization * stepsize * iterationNumber)^(-3/4)
> See also how to properly select the gains [1].
> Resources:
> [1] http://arxiv.org/pdf/1107.2490.pdf
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
