[
https://issues.apache.org/jira/browse/SYSTEMML-845?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15408502#comment-15408502
]
Mike Dusenberry commented on SYSTEMML-845:
------------------------------------------
[~niketanpansare] When running in forced singlenode mode, these two scripts
have equal performance (time & accuracy). However, I'm seeing severe
regressions when attempting to run with Spark (--master local[*], 50G driver
memory), or even just standalone mode with hybrid-mr mode. For example, I
limited the {{epochs}} in both scripts to just 2, and then ran with
{{-explain}} and {{-stats}}, and it showed a very large number of Spark jobs
(as well as MR jobs in the hybrid-mr mode). The logs are too large to include.
Looking through the explain plan, it looks like many operations are being
mapped to Spark jobs in both scripts. Any thoughts?
> Compare Performance of LeNet Scripts With & Without Using SystemML-NN
> ---------------------------------------------------------------------
>
> Key: SYSTEMML-845
> URL: https://issues.apache.org/jira/browse/SYSTEMML-845
> Project: SystemML
> Issue Type: Improvement
> Reporter: Mike Dusenberry
> Attachments: convert.dml, log08.03.16-1470268602.txt, perf.sh, run.sh
>
>
> This JIRA issue tracks the comparison of the performance of the LeNet scripts
> with & without using SystemML-NN. The goal is that they should have equal
> performance in terms of both accuracy and time. Any difference will be
> indicate areas of engine improvement.
> Scripts:
> * [mnist_lenet-train.dml |
> https://github.com/apache/incubator-systemml/blob/master/scripts/staging/SystemML-NN/examples/mnist_lenet-train.dml]
> - LeNet script that *does* use the SystemML-NN library.
> * [lenet-train.dml |
> https://github.com/apache/incubator-systemml/blob/master/scripts/staging/lenet-train.dml]
> - LeNet script that *does not* use the SystemML-NN library.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
