Mailing-List: contact issues-help@flink.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@flink.apache.org
Date: Fri, 22 Apr 2016 13:41:12 +0000 (UTC)
From: "ASF GitHub Bot (JIRA)" <jira@apache.org>
To: issues@flink.apache.org
Message-ID: <JIRA.12839550.1434979487000.8418.1461332472872@Atlassian.JIRA>
In-Reply-To: <JIRA.12839550.1434979487000@Atlassian.JIRA>
References: <JIRA.12839550.1434979487000@Atlassian.JIRA>
 <JIRA.12839550.1434979487648@arcas>
Subject: [jira] [Commented] (FLINK-2259) Support training Estimators using a
 (train, validation, test) split of the available data
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/FLINK-2259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15253944#comment-15253944 ] 

ASF GitHub Bot commented on FLINK-2259:
---------------------------------------

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

    https://github.com/apache/flink/pull/1898#discussion_r60739883
  
    --- Diff: flink-libraries/flink-ml/src/test/scala/org/apache/flink/ml/preprocessing/SplitterITSuite.scala ---
    @@ -0,0 +1,73 @@
    +/*
    + * 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.preprocessing
    +
    +import org.apache.flink.api.scala.ExecutionEnvironment
    +import org.apache.flink.api.scala._
    +import org.apache.flink.test.util.FlinkTestBase
    +import org.scalatest.{Matchers, FlatSpec}
    +import org.apache.flink.ml.math.Vector
    +import org.apache.flink.api.scala.utils._
    +
    +
    +class SplitterITSuite extends FlatSpec
    +  with Matchers
    +  with FlinkTestBase {
    +
    +  behavior of "Flink's DataSet Splitter"
    +
    +  import MinMaxScalerData._
    +
    + it should "result in datasets with no elements in common and all elements used" in {
    +    val env = ExecutionEnvironment.getExecutionEnvironment
    +
    +    val dataSet = env.fromCollection(data)
    +
    +    val splitDataSets = Splitter.randomSplit(dataSet.zipWithIndex, 0.5)
    +
    +   (splitDataSets(0).count() + splitDataSets(1).count()) should equal(dataSet.count())
    +
    +
    +   splitDataSets(0).join(splitDataSets(1)).where(0).equalTo(0).count() should equal(0)
    +  }
    +
    +  it should "result in datasets of an expected size when precise" in {
    +    val env = ExecutionEnvironment.getExecutionEnvironment
    +
    +    val dataSet = env.fromCollection(data)
    +
    +    val splitDataSets = Splitter.randomSplit(dataSet, 0.5)
    +
    +    val expectedLength = dataSet.count().toDouble * 0.5
    +
    +    splitDataSets(0).count().toDouble should equal(expectedLength +- 5.0)
    +  }
    +
    +  it should "result in expected number of datasets" in {
    +    val env = ExecutionEnvironment.getExecutionEnvironment
    +
    +    val dataSet = env.fromCollection(data)
    +
    +    val fracArray = Array(0.5, 0.25, 0.25)
    +
    +    val splitDataSets = Splitter.multiRandomSplit(dataSet, fracArray)
    +
    +    splitDataSets.length should equal(fracArray.length)
    +  }
    +
    --- End diff --
    
    Maybe we could add a test case for sampling with replacement?


> Support training Estimators using a (train, validation, test) split of the available data
> -----------------------------------------------------------------------------------------
>
>                 Key: FLINK-2259
>                 URL: https://issues.apache.org/jira/browse/FLINK-2259
>             Project: Flink
>          Issue Type: New Feature
>          Components: Machine Learning Library
>            Reporter: Theodore Vasiloudis
>            Assignee: Trevor Grant
>            Priority: Minor
>              Labels: ML
>
> When there is an abundance of data available, a good way to train models is to split the available data into 3 parts: Train, Validation and Test.
> We use the Train data to train the model, the Validation part is used to estimate the test error and select hyperparameters, and the Test is used to evaluate the performance of the model, and assess its generalization [1]
> This is a common approach when training Artificial Neural Networks, and a good strategy to choose in data-rich environments. Therefore we should have some support of this data-analysis process in our Estimators.
> [1] Friedman, Jerome, Trevor Hastie, and Robert Tibshirani. The elements of statistical learning. Vol. 1. Springer, Berlin: Springer series in statistics, 2001.


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