[ https://issues.apache.org/jira/browse/SPARK-7148?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14594261#comment-14594261 ]
Joseph K. Bradley commented on SPARK-7148: ------------------------------------------ Hm, if it's that simple, then I wonder if we can adjust parquet.block.size before saving/loading the ML models and reset the block size to its original value afterwards. I'll have to try that! > Configure Parquet block size (row group size) for ML model import/export > ------------------------------------------------------------------------ > > Key: SPARK-7148 > URL: https://issues.apache.org/jira/browse/SPARK-7148 > Project: Spark > Issue Type: Improvement > Components: MLlib, SQL > Affects Versions: 1.3.0, 1.3.1, 1.4.0 > Reporter: Joseph K. Bradley > Priority: Minor > > It would be nice if we could configure the Parquet buffer size when using > Parquet format for ML model import/export. Currently, for some models (trees > and ensembles), the schema has 13+ columns. With a default buffer size of > 128MB (I think), that puts the allocated buffer way over the default memory > made available by run-example. Because of this problem, users have to use > spark-submit and explicitly use a larger amount of memory in order to run > some ML examples. > Is there a simple way to specify {{parquet.block.size}}? I'm not familiar > with this part of SparkSQL. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org