[
https://issues.apache.org/jira/browse/SPARK-59313?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ruifeng Zheng updated SPARK-59313:
----------------------------------
Description: FeedForwardModel lazily allocates outputs and deltas as
reusable scratch buffers for forward and gradient computations. Mark them
transient because they can be recreated after deserialization and otherwise
cached batch matrices are unnecessarily serialized with the model.
Summary: Avoid serializing FeedForwardModel scratch buffers (was:
Eagerly initialize the ANN model in MultilayerPerceptronClassificationModel)
> Avoid serializing FeedForwardModel scratch buffers
> --------------------------------------------------
>
> Key: SPARK-59313
> URL: https://issues.apache.org/jira/browse/SPARK-59313
> Project: Spark
> Issue Type: Sub-task
> Components: ML
> Affects Versions: 5.0.0
> Reporter: Ruifeng Zheng
> Priority: Major
> Labels: pull-request-available
>
> FeedForwardModel lazily allocates outputs and deltas as reusable scratch
> buffers for forward and gradient computations. Mark them transient because
> they can be recreated after deserialization and otherwise cached batch
> matrices are unnecessarily serialized with the model.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
