GitHub user viirya opened a pull request:
https://github.com/apache/spark/pull/2217
[SPARK-3327] Make broadcasted value mutable for caching useful information
This PR makes broadcasted value mutable for caching useful information when
implementing some algorithms that iteratively run with those information.
Some algorithms can be implemented more efficiently if we can update
broadcasted variables and use them in later iterations.
Specifically, we would like to performa operation "A" on each partition of
data. Some variables are updated. Then we want to run operation "B" on the data
too. "B" operation uses the variables updated by operation "A".
One of the examples is the Liblinear on Spark from Dr. Lin. They discuss
the problem in Section IV.D of the paper "Large-scale Logistic Regression and
Linear Support Vector Machines Using Spark."
This implementation can make sure that updated broadcasted variables are
consistent after parallel data operations (such as map) are done on the
broadcasted variables at same node.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/viirya/spark-1 mutable_broadcast
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/2217.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #2217
----
commit 274b7daccb9986e1de68a6669a8893d3a3818cd0
Author: Liang-Chi Hsieh <[email protected]>
Date: 2014-08-30T18:37:41Z
Make broadcasted value mutable for caching useful information when
implementing some algorithms.
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
