You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@predictionio.apache.org by "Mars Hall (JIRA)" <ji...@apache.org> on 2017/07/15 00:07:01 UTC
[jira] [Updated] (PIO-105) Batch Predictions
[ https://issues.apache.org/jira/browse/PIO-105?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Mars Hall updated PIO-105:
--------------------------
External issue URL: https://github.com/apache/incubator-predictionio/pull/412
Description:
Implement a new {{pio batchpredict}} command to enable massive, fast, batch predictions from a trained model. Read a multi-object JSON file as the input format, with one query object per line. Similarly, write results to a multi-object JSON file, with one prediction result + its original query per line.
Currently getting bulk predictions from PredictionIO is possible with either:
* a {{pio eval}} script, which will always train a fresh, unvalidated model before getting predictions
* a custom script that hits the {{queries.json}} HTTP API, which is a serious bottleneck when requesting hundreds-of-thousands or millions of predictions
Neither of these existing bulk-prediction hacks are adequate for the reasons mentioned.
It's time for this use-case to be a firstclass command :D
Pull request https://github.com/apache/incubator-predictionio/pull/412
was:
Implement a new {{pio batchpredict}} command to enable massive, fast, batch predictions from a trained model. Read a multi-object JSON file as the input format, with one query object per line. Similarly, write results to a multi-object JSON file, with one prediction result + its original query per line.
Currently getting bulk predictions from PredictionIO is possible with either:
* a {{pio eval}} script, which will always train a fresh, unvalidated model before getting predictions
* a custom script that hits the {{queries.json}} HTTP API, which is a serious bottleneck when requesting hundreds-of-thousands or millions of predictions
Neither of these existing bulk-prediction hacks are adequate for the reasons mentioned.
It's time for this use-case to be a firstclass command :D
> Batch Predictions
> -----------------
>
> Key: PIO-105
> URL: https://issues.apache.org/jira/browse/PIO-105
> Project: PredictionIO
> Issue Type: New Feature
> Components: Core
> Reporter: Mars Hall
> Assignee: Mars Hall
>
> Implement a new {{pio batchpredict}} command to enable massive, fast, batch predictions from a trained model. Read a multi-object JSON file as the input format, with one query object per line. Similarly, write results to a multi-object JSON file, with one prediction result + its original query per line.
> Currently getting bulk predictions from PredictionIO is possible with either:
> * a {{pio eval}} script, which will always train a fresh, unvalidated model before getting predictions
> * a custom script that hits the {{queries.json}} HTTP API, which is a serious bottleneck when requesting hundreds-of-thousands or millions of predictions
> Neither of these existing bulk-prediction hacks are adequate for the reasons mentioned.
> It's time for this use-case to be a firstclass command :D
> Pull request https://github.com/apache/incubator-predictionio/pull/412
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)