zhengruifeng removed a comment on issue #27145: [SPARK-30457][ML][WIP] Use PeriodicRDDCheckpointer instead of NodeIdCache URL: https://github.com/apache/spark/pull/27145#issuecomment-572395974 I made this change mainly to reduce redundant impls, however, after I test the performance, I guess there maybe some recomputation issue in `NodeIdCache`: test code: ```scala import org.apache.spark.ml.regression._ import org.apache.spark.storage.StorageLevel var df = spark.read.format("libsvm").load("/data1/Datasets/a9a/a9a") (0 until 8).foreach{ _ => df = df.union(df) } df.persist(StorageLevel.MEMORY_AND_DISK) df.count df.count df.count val start = System.currentTimeMillis; val gbtm = new GBTRegressor().setCacheNodeIds(true).fit(df); val end = System.currentTimeMillis; end - start val start = System.currentTimeMillis; val rfm = new RandomForestRegressor().setCacheNodeIds(true).fit(df); val end = System.currentTimeMillis; end - start ``` Durations: using Checkpointer: GBT:143342, RF:28081 using NodeIdCache: GBT:333915, RF:60631 `PeriodicRDDCheckpointer` will cached one more rdd than `NodeIdCache`, I am not sure whether this makes the change.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
