[
https://issues.apache.org/jira/browse/SPARK-10182?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14709416#comment-14709416
]
Vyacheslav Baranov commented on SPARK-10182:
--------------------------------------------
Sorry, it was wrong spark-shell. Behaviour in master is slightly different: It
looks like RDDs are removed from cache on GC. I had to modify the code a bit to
reproduce the issue:
{code}
import org.apache.spark.SparkContext
import org.apache.spark.mllib.classification.LogisticRegressionWithLBFGS
import org.apache.spark.mllib.linalg.Vectors
import org.apache.spark.mllib.regression.LabeledPoint
for (i <- 0 until 100) {
val samples = Seq[LabeledPoint](
LabeledPoint(1.0, Vectors.dense(1.0, 0.0)),
LabeledPoint(1.0, Vectors.dense(0.0, 1.0)),
LabeledPoint(0.0, Vectors.dense(1.0, 1.0)),
LabeledPoint(0.0, Vectors.dense(0.0, 0.0))
)
val rdd = sc.parallelize(samples)
val model = {
new LogisticRegressionWithLBFGS()
.setNumClasses(2)
.run(rdd)
.clearThreshold()
}
}
{code}
!http://piqqin.com/img/ea6c54a1bf414828a794ca6604436d78.png!
The number of cached RDDs decreases over time. However, on real-size data when
building cross-validated models this is real problem: useful pre-cached
datasets are dropped from memory and replaced with these
{{MapPartitionsRDD}}'s. With the fix I've submitted behaviour is perfectly
fine: only one RDD is cached at a time, so pre-cached data is untouched.
> GeneralizedLinearModel doesn't unpersist cached data
> ----------------------------------------------------
>
> Key: SPARK-10182
> URL: https://issues.apache.org/jira/browse/SPARK-10182
> Project: Spark
> Issue Type: Bug
> Components: MLlib
> Affects Versions: 1.4.1
> Reporter: Vyacheslav Baranov
>
> The problem might be reproduced in spark-shell with following code snippet:
> {code}
> import org.apache.spark.SparkContext
> import org.apache.spark.mllib.classification.LogisticRegressionWithLBFGS
> import org.apache.spark.mllib.linalg.Vectors
> import org.apache.spark.mllib.regression.LabeledPoint
> val samples = Seq[LabeledPoint](
> LabeledPoint(1.0, Vectors.dense(1.0, 0.0)),
> LabeledPoint(1.0, Vectors.dense(0.0, 1.0)),
> LabeledPoint(0.0, Vectors.dense(1.0, 1.0)),
> LabeledPoint(0.0, Vectors.dense(0.0, 0.0))
> )
> val rdd = sc.parallelize(samples)
> for (i <- 0 until 10) {
> val model = {
> new LogisticRegressionWithLBFGS()
> .setNumClasses(2)
> .run(rdd)
> .clearThreshold()
> }
> }
> {code}
> After code execution there are 10 {{MapPartitionsRDD}} objects on "Storage"
> tab in Spark application UI.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
