[
https://issues.apache.org/jira/browse/SPARK-29832?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16975799#comment-16975799
]
Sean R. Owen commented on SPARK-29832:
--------------------------------------
[~spark_cachecheck] some of these may be valid, but a lot of them don't appear
to be. Let's tackle a few before opening the 30 JIRAs you did -- that's very
noisy.
run() is going to do use (a transform of) this input many times in a loop. I
don't think this analysis is accurate. You could say that it's more optimal to
persist stuff closer to where the loop is; sometimes it's better, sometimes,
not, depends on how big and expensive the result is.
> Unnecessary persist on instances in ml.regression.IsotonicRegression.fit
> ------------------------------------------------------------------------
>
> Key: SPARK-29832
> URL: https://issues.apache.org/jira/browse/SPARK-29832
> Project: Spark
> Issue Type: Improvement
> Components: ML
> Affects Versions: 3.0.0
> Reporter: Dong Wang
> Priority: Major
>
> Persist on instances in ml.regression.IsotonicRegression.fit() is
> unnecessary, because it is only used once in run(instances).
> {code:scala}
> override def fit(dataset: Dataset[_]): IsotonicRegressionModel =
> instrumented { instr =>
> transformSchema(dataset.schema, logging = true)
> // Extract columns from data. If dataset is persisted, do not persist
> oldDataset.
> val instances = extractWeightedLabeledPoints(dataset)
> val handlePersistence = dataset.storageLevel == StorageLevel.NONE
> // Unnecessary persist
> if (handlePersistence) instances.persist(StorageLevel.MEMORY_AND_DISK)
> instr.logPipelineStage(this)
> instr.logDataset(dataset)
> instr.logParams(this, labelCol, featuresCol, weightCol, predictionCol,
> featureIndex, isotonic)
> instr.logNumFeatures(1)
> val isotonicRegression = new
> MLlibIsotonicRegression().setIsotonic($(isotonic))
> val oldModel = isotonicRegression.run(instances) // Only use once here
> if (handlePersistence) instances.unpersist()
> {code}
> This issue is reported by our tool CacheCheck, which is used to dynamically
> detecting persist()/unpersist() api misuses.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
