How to build a distributed recommender with Graphlab Create

User 292 | 5/6/2014, 6:26:52 AM

I have 10 machines and I want to build a distributed recommender with Graphlab Create, is it practicable? I'll appreciate for your help.


User 6 | 5/6/2014, 3:17:04 PM

Hi, GraphLab Create does not support a distributed cluster yet, it is on our roadmap, as you can see here: Currently GraphLab Create runs on a single machine (or using a single ec2 instance spawn by your machine).

In the meantime you can use GraphLab open source in case you want to utilize 10 machines.

User 292 | 5/8/2014, 5:58:34 AM

Thank you very much, your answer is of great help to me.

User 318 | 5/28/2014, 9:49:44 AM

I see that HDFS connector is supported. Does it mean if I build a recommender with Graphlab Create which reads data from HDFS hosted on a distributed cluster, it is running in a distributed manner?

User 6 | 5/28/2014, 5:21:34 PM

No, it means that in GraphLab Create we read the data from HDFS, process it on a single multicore machine and write it back to HDFS.

The open source GraphLab does support parallel (cluster) execution and reading and writing to HDFS.