Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/1898#discussion_r63499292
--- Diff:
flink-libraries/flink-ml/src/main/scala/org/apache/flink/ml/preprocessing/Splitter.scala
---
@@ -0,0 +1,210 @@
+/*
+ * 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._
+import org.apache.flink.api.scala.DataSet
+import org.apache.flink.api.scala.utils._
+
+
+import org.apache.flink.ml.common.{FlinkMLTools, ParameterMap,
WithParameters}
+import org.apache.flink.util.Collector
+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] 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.zipWithUniqueId
+
+ if ((fraction >= 1) || (fraction <= 0)) {
+ throw new IllegalArgumentException("sampling fraction outside of
(0,1) bounds is nonsensical")
+ }
+
+ val leftSplit: DataSet[(Long, T)] = precise match {
+ case false => indexedInput.sample(false, fraction, seed)
+ case true => {
+ val count = indexedInput.count()
--- End diff --
The `count` operation can be quite expensive since we're triggering the
execution of the flink job at this point. At the moment, the `sampleWithSize`
methods requires a number of samples as the parameter. Maybe we could extend
the functionality in the future so that one can also give a fraction value
instead. Then the method could calculate the count without the `count` method
by simply doing a reduce operation and then broadcasting the result to all
sample operators.
So maybe we could add a comment that this could/should be improved once we
have this functionality.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
