5

I have created data sets of various sizes say 1GB, 2GB, 3GB, 4GB (< 10 GB) and executing various machine learning models on Azure ML.

1) Can I know what is the server specifications (RAM, CPU) that is provided in the Azure ML service.

2) Also at times the reader says "Memory exhaust" for >4GB of data.Though azure ml should be able to handle 10GB of data as per documentation.

3) If I run multiple experiments(in different tabs of browser) in parallel, its taking more time.

4) Is there any way to set the RAM, CPU cores in Azure ML

marnun
  • 808
  • 8
  • 23
Naseer
  • 51
  • 3
  • How about a screen shot of your experiment? – Mike Wise Feb 25 '16 at 09:32
  • Are you using a free workspace or paid? – GregGalloway Feb 25 '16 at 10:01
  • 1
    in AzureML training your data is FREE! We do not post the specs for the VMs that we use, as they may change. We run everything inside of equal containers. Experiment execution can not be used to determine timing of a web service though, which I think is more important. Check this out to understand the timing behind a web service call. http://stackoverflow.com/questions/34990561/azure-machine-learning-request-response-latency?rq=1 – Dan Ciborowski - MSFT Mar 08 '16 at 14:16
  • That isn't 100% true. Microsoft used to post the VM specs for Azure ML Studio. If you can find them in search engine cache it was listed as an A7 with 56gig RAM. As @DanCiborowski-MSFT mentioned these VMs can change. However, the core underlying CPU/memory footprint has stayed the same. In fact, you can run a simply Python/R script to verify it is indeed still 56 gig RAM (as of May 2019). – Bart Czernicki May 11 '19 at 19:07

2 Answers2

2

I have a partial answer: 1. no, it's abstracted

  1. The following types of data can expand into larger datasets during feature normalization, and are limited to less than 10 GB:

    Sparse Categorical Strings Binary data

(see this)

  1. I'm not sure, but while working on it, I didn't experience any change when running a single experiment and multiple experiment

  2. you can scale the machines in the standard tier (see this)

marnun
  • 808
  • 8
  • 23
0

I would recommend looking at the new "Visual Interface" for Azure ML service, which allows you to go well over the 10gig limit and bring your own compute clusters.

//BUILD 2019 announcement video: https://www.youtube.com/watch?v=QBPCaZo9xx0

Bart Czernicki
  • 3,663
  • 1
  • 21
  • 19