Predictive Services

User 1456 | 3/11/2015, 8:06:36 AM

Currently the predictive services API usage is restricted to EC2 , m3.large machine. Is there any way to use it on a local enviroment? Internally does it use node.js for binding??


User 398 | 3/11/2015, 5:46:26 PM

Hi Mahesh,

Thanks for using GraphLab Create. Predictive Services are restricted to EC2 instances of type m3.large or better. (We also support some of the older instance types such as m1.large+ and m2.large+.) At the moment, there is no way to launch them locally. Can you tell me a little bit more about your use case? Are you trying to launch on your own server or on your local workstation (for development reasons)? We are working to allow on-premise deployments, but we're not there yet.

In regards to your question about node.js, I'm not totally sure what you mean by "binding". We are fans of node.js, but we are not using it within Predictive Services. Why do you ask?


User 1456 | 3/13/2015, 2:02:46 PM

Thanks for the reply Robert. Yes, as you said i am looking for some on-premise kinda deplyoment for development purpose. I was wondering if there was a possibility of developing predictive service kind of api's using node.js or build kind of simple python server like flask that can wrap Graph lab create.

User 398 | 3/13/2015, 5:21:16 PM

Hi Mahesh --

What you are describing sounds reasonable and is in fact similar to what we are doing already for Predictive Services. Using node.js would be tricky since there is no JavaScript client to GraphLab Create. It's certainly possible to wrap GraphLab Create in a Flask webservice, however. With Predictive Services you get a few extra bells and whistles though, such as load balancing, distributed caching, metrics, monitoring, and you can resize your service with a single API call. We are actively working on supporting on premise deployments and would love to hear more about what you're looking for. Feel free to send me an email (robert at dato dot com).

Thanks, Robert