[
https://issues.apache.org/jira/browse/FLINK-1723?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14618613#comment-14618613
]
ASF GitHub Bot commented on FLINK-1723:
---------------------------------------
Github user thvasilo commented on a diff in the pull request:
https://github.com/apache/flink/pull/891#discussion_r34148875
--- Diff:
flink-staging/flink-ml/src/main/scala/org/apache/flink/ml/evaluation/CrossValidation.scala
---
@@ -0,0 +1,97 @@
+/*
+ * 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.evaluation
+
+import org.apache.flink.api.scala._
+import org.apache.flink.ml.RichDataSet
+import java.util.Random
+
+import org.apache.flink.ml.pipeline.{EvaluateDataSetOperation,
FitOperation, Predictor}
+
+object CrossValidation {
+ def crossValScore[P <: Predictor[P], T](
+ predictor: P,
+ data: DataSet[T],
+ scorerOption: Option[Scorer] = None,
+ cv: FoldGenerator = KFold(),
+ seed: Long = new Random().nextLong())(implicit fitOperation:
FitOperation[P, T],
+ evaluateDataSetOperation: EvaluateDataSetOperation[P, T, Double]):
Array[DataSet[Double]] = {
+ val folds = cv.folds(data, 1)
+
+ val scores = folds.map {
+ case (training: DataSet[T], testing: DataSet[T]) =>
+ predictor.fit(training)
+ if (scorerOption.isEmpty) {
+ predictor.score(testing)
+ } else {
+ val s = scorerOption.get
+ s.evaluate(testing, predictor)
+ }
+ }
+ // TODO: Undecided on the return type: Array[DS[Double]] or DS[Double]
i.e. reduce->union?
+ // Or: Return mean and std?
+ scores//.reduce((right: DataSet[Double], left: DataSet[Double]) =>
left.union(right)).mean()
--- End diff --
sklearn returns an array of scores, as does spark. The mean on its own does
not tell us much, it is very useful to see how much variance the scores have as
well, i.e. if we have a robust model or not. Leaving it up to the user I think
is a good choice here, so I would vote for returning a DataSet[Double], with
all the scores (i.e. reduce->union on the scores array).
> Add cross validation for model evaluation
> -----------------------------------------
>
> Key: FLINK-1723
> URL: https://issues.apache.org/jira/browse/FLINK-1723
> Project: Flink
> Issue Type: New Feature
> Components: Machine Learning Library
> Reporter: Till Rohrmann
> Assignee: Theodore Vasiloudis
> Labels: ML
>
> Cross validation [1] is a standard tool to estimate the test error for a
> model. As such it is a crucial tool for every machine learning library.
> The cross validation should work with arbitrary Estimators and error metrics.
> A first cross validation strategy it should support is the k-fold cross
> validation.
> Resources:
> [1] [http://en.wikipedia.org/wiki/Cross-validation]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
