[
https://issues.apache.org/jira/browse/SPARK-3728?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15381583#comment-15381583
]
Xusen Yin commented on SPARK-3728:
----------------------------------
Not now. Because I thought the BFS style could reach the best parallelism,
while the DFS may harm the parallel ability. And IMHO the BFS style training is
not the root cause of out-of-memory during the training phase of RandomForest.
Do you have any suggestions on this?
> RandomForest: Learn models too large to store in memory
> -------------------------------------------------------
>
> Key: SPARK-3728
> URL: https://issues.apache.org/jira/browse/SPARK-3728
> Project: Spark
> Issue Type: Improvement
> Components: MLlib
> Reporter: Joseph K. Bradley
>
> Proposal: Write trees to disk as they are learned.
> RandomForest currently uses a FIFO queue, which means training all trees at
> once via breadth-first search. Using a FILO queue would encourage the code
> to finish one tree before moving on to new ones. This would allow the code
> to write trees to disk as they are learned.
> Note: It would also be possible to write nodes to disk as they are learned
> using a FIFO queue, once the example--node mapping is cached [JIRA]. The
> [Sequoia Forest package]() does this. However, it could be useful to learn
> trees progressively, so that future functionality such as early stopping
> (training fewer trees than expected) could be supported.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
