[ https://issues.apache.org/jira/browse/FLINK-2259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15253943#comment-15253943 ]
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_r60739818 --- Diff: flink-libraries/flink-ml/src/main/scala/org/apache/flink/ml/preprocessing/Splitter.scala --- @@ -0,0 +1,215 @@ +/* + * 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.common.typeinfo.{TypeInformation, BasicTypeInfo} +import org.apache.flink.api.java.Utils +import org.apache.flink.api.scala. DataSet +import org.apache.flink.api.scala.utils._ + +import org.apache.flink.ml.common.{FlinkMLTools, ParameterMap, WithParameters} +import _root_.scala.reflect.ClassTag + +object Splitter { + + case class TrainTestDataSet[T: TypeInformation : ClassTag](training: DataSet[T], + testing: DataSet[T]) + + case class TrainTestHoldoutDataSet[T: TypeInformation : ClassTag](training: DataSet[T], + testing: DataSet[T], + holdout: DataSet[T]) + // -------------------------------------------------------------------------------------------- + // randomSplit + // -------------------------------------------------------------------------------------------- + /** + * Split a DataSet by the probability fraction of each element. + * + * @param input DataSet to be split + * @param fraction Probability that each element is chosen, should be [0,1] without + * replacement, and [0, ∞) with replacement. While fraction is larger + * than 1, the elements are expected to be selected multi times into + * sample on average. This fraction refers to the first element in the + * resulting array. + * @param precise Sampling by default is random and can result in slightly lop-sided + * sample sets. When precise is true, equal sample set size are forced, + * however this is somewhat less efficient. + * @param seed Random number generator seed. + * @return An array of two datasets + */ + + def randomSplit[T: TypeInformation : ClassTag]( input: DataSet[T], + fraction: Double, + precise: Boolean = false, + seed: Long = Utils.RNG.nextLong()) + : Array[DataSet[T]] = { + import org.apache.flink.api.scala._ + + val indexedInput: DataSet[(Long, T)] = input.zipWithIndex + + val leftSplit: DataSet[(Long, T)] = precise match { + case false => indexedInput.sample(false, fraction, seed) --- End diff -- What happens if fraction is larger than `1` and `withReplacement` is set to `false`? Shouldn't it be set to `true` in this case? > 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)