[
https://issues.apache.org/jira/browse/MAHOUT-1416?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13899184#comment-13899184
]
Manoj Awasthi edited comment on MAHOUT-1416 at 2/12/14 3:37 PM:
----------------------------------------------------------------
[~smarthi] I agree with you. I considered that but was thinking of keeping the
"load forest & classify" semantics of usage while proposing this patch. But
now, I also think "load" would not be an appropriate name for a read from
stream. Thanks for the comment.
was (Author: awmanoj):
[~smarthi] I agree with you. I considered that but was thinking of keeping the
"load forest & classify" semantics of usage while proposing this patch. Now
while reading your comment, although, I am more inclined towards just making
the read() function public. What do you think of it?
> Implement an interface to load a forest from an in-memory byte buffer
> ---------------------------------------------------------------------
>
> Key: MAHOUT-1416
> URL: https://issues.apache.org/jira/browse/MAHOUT-1416
> Project: Mahout
> Issue Type: Improvement
> Affects Versions: 1.0
> Reporter: Manoj Awasthi
> Fix For: 1.0
>
> Attachments: MAHOUT-1416.patch
>
> Original Estimate: 24h
> Remaining Estimate: 24h
>
> There is a use case I encounter in which a random forest model is passed to
> remote server over a REST API and is used for scoring the incoming requests
> in there.
> In such a scenario it is highly inefficient to write the byte buffer received
> at the remote server on the filesystem and then use Mahout APIs to load the
> forest.
> We should be able to load the forest from the byte buffer received. This is
> useful in non-map-reduce usecases of classification like that mentioned above.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
