[
https://issues.apache.org/jira/browse/PIO-38?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15523861#comment-15523861
]
Wojciech Indyk commented on PIO-38:
-----------------------------------
Hello [~Ziemin]! Sorry for late response.
I would like to have a chance to provide events to PredictionIO using my
current place of storing events. As I can see PredictionIO can work with a pair
of Elasticsearch+HBase. Therefore to use Elasticsearch as a backend I need to
use HBase as an event-store. I don't know PredictionIO so good, so correct me
if I'm wrong.
I don't want to use HBase, because it enlarges my technology stack and has no
benefit in case of training model in batch. Parquet is more suitable to this
case, when I append my archive of events once a day, then can use this data
(subset) to train a recommendation model without duplication data in HBase.
Is it clear enough?
> add Apache Parquet as a data source
> -----------------------------------
>
> Key: PIO-38
> URL: https://issues.apache.org/jira/browse/PIO-38
> Project: PredictionIO
> Issue Type: New Feature
> Reporter: Wojciech Indyk
> Labels: features
>
> Apache Parquet (https://parquet.apache.org/) is a columnar data store, native
> for Apache Spark and very well suited to storing batch data (as an input) for
> PredictionIO Engine.
> Parquet is very popular to archive clickstream, so it would enable to use
> PredictionIO without additional import of data (and duplication) to HBase.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
