flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-1807) Stochastic gradient descent optimizer for ML library
Date Wed, 29 Apr 2015 15:40:06 GMT

    [ https://issues.apache.org/jira/browse/FLINK-1807?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14519585#comment-14519585
] 

ASF GitHub Bot commented on FLINK-1807:
---------------------------------------

Github user tillrohrmann commented on a diff in the pull request:

    https://github.com/apache/flink/pull/613#discussion_r29348422
  
    --- Diff: flink-staging/flink-ml/src/main/scala/org/apache/flink/ml/optimization/RegularizationType.scala
---
    @@ -0,0 +1,116 @@
    +/*
    + * Licensed to the Apache Software Foundation (ASF) under one
    + * or more contributor license agreements.  See the NOTICE file
    + * distributed with this work for additional information
    + * regarding copyright ownership.  The ASF licenses this file
    + * to you under the Apache License, Version 2.0 (the
    + * "License"); you may not use this file except in compliance
    + * with the License.  You may obtain a copy of the License at
    + *
    + *     http://www.apache.org/licenses/LICENSE-2.0
    + *
    + * Unless required by applicable law or agreed to in writing, software
    + * distributed under the License is distributed on an "AS IS" BASIS,
    + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    + * See the License for the specific language governing permissions and
    + * limitations under the License.
    + */
    +
    +package org.apache.flink.ml.optimization
    +
    +import breeze.numerics._
    +import org.apache.flink.ml.math.{BLAS, Vector}
    +import org.apache.flink.ml.math.Breeze._
    +import breeze.linalg.{norm => BreezeNorm, Vector => BreezeVector, max}
    +
    +// TODO(tvas): Change name to RegularizationPenalty?
    +abstract class RegularizationType extends Serializable{
    +  /** Calculates and applies the regularization amount and the regularization parameter
    +    *
    +    * @param weights The old weights
    +    * @param effectiveStepSize The effective step size for this iteration
    +    * @param regParameter The current regularization parameter
    +    * @return A tuple whose first element is the updated weight vector and the second
is the
    +    *         new regularization parameter
    +    */
    +  def applyRegularization(weights: Vector, effectiveStepSize: Double, regParameter: Double):
    +  (Vector, Double)
    +  // TODO(tvas): We are not currently using the regularization value anywhere, but it
could be
    +  // useful to keep a history of it.
    +
    +}
    +
    +class NoRegularization extends RegularizationType {
    +  /** Returns the original weights without any regularization applied
    +    *
    +    * @param weights The old weights
    +    * @param effectiveStepSize The effective step size for this iteration
    +    * @param regParameter The current regularization parameter
    +    * @return A tuple whose first element is the updated weight vector and the second
is the
    +    *         regularization value
    +    */
    +  override def applyRegularization(weights: Vector,
    +                                   effectiveStepSize: Double,
    +                                   regParameter: Double):
    +  (Vector, Double) = {(weights, 0.0)}
    +}
    +
    +class L2Regularization extends RegularizationType {
    +  /** Calculates and applies the regularization amount and the regularization parameter
    +    *
    +    * Implementation was taken from the Apache Spark Mllib library:
    +    * http://git.io/vfZIT
    +    * @param weights The old weights
    +    * @param effectiveStepSize The effective step size for this iteration
    +    * @param regParameter The current regularization parameter
    +    * @return A tuple whose first element is the updated weight vector and the second
is the
    +    *         regularization value
    +    */
    +  override def applyRegularization(weights: Vector,
    +                                   effectiveStepSize: Double,
    +                                   regParameter: Double):
    +  (Vector, Double) = {
    +
    +    val brzWeights: BreezeVector[Double] = weights.asBreeze
    +    brzWeights :*= (1.0 - effectiveStepSize * regParameter)
    --- End diff --
    
    I think you have to apply the regularization gradient on the old weight vector and not
on the new weight vector to make it mathematically sound here.


> Stochastic gradient descent optimizer for ML library
> ----------------------------------------------------
>
>                 Key: FLINK-1807
>                 URL: https://issues.apache.org/jira/browse/FLINK-1807
>             Project: Flink
>          Issue Type: Improvement
>          Components: Machine Learning Library
>            Reporter: Till Rohrmann
>            Assignee: Theodore Vasiloudis
>              Labels: ML
>
> Stochastic gradient descent (SGD) is a widely used optimization technique in different
ML algorithms. Thus, it would be helpful to provide a generalized SGD implementation which
can be instantiated with the respective gradient computation. Such a building block would
make the development of future algorithms easier.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message