none
Overflow in Machine Learning experiment RRS feed

  • Question

  • Hi all,

    I'm currently running an experiment on ML Azure using Open Class Support Vector Machine, Train Anomaly Detection and a csv corpus. The experiments failed after almost 12 minutes of running. The message error that I get is the following : 

    requestId = 07e1b8d3-8276-4461-b76c-d3c32bb70fa8 errorComponent=Module. taskStatusCode=500. {"Exception":{"ErrorId":"InternalError","ErrorCode":"0000","ExceptionType":"ModuleException","Message":"Sorry, it seems that you have encountered an internal system error. Please contact amlforum@microsoft.com with the full URL in the browser and the time you experienced the failure. We can locate this error with your help and investigate further. Thank you.","Exception":{"ExceptionType":"Exception","Message":"Exception has been thrown by the target of an invocation.","Exception":{"ExceptionType":"Exception","Message":"Arithmetic operation resulted in an overflow."}}}}Error: Sorry, it seems that you have encountered an internal system error. Please contact amlforum@microsoft.com with the full URL in the browser and the time you experienced the failure. We can locate this error with your help and investigate further. Thank you. Process exited with error code -2

    P-S : I've already sent a mail to amlforum a while ago but didn't receive an answer for them, so any help will be more than welcome :).

    Thursday, June 23, 2016 11:04 AM

Answers

  • Hey ImaneR!

    Sorry about that, we've actually been discussing this a bit internally to try to solve this issue and have come to the conclusion that we'd need a more specific repro. The fundamental error comes from creating the support vectors, and is data-specific.

    If your data doesn't contain any PII or other sensitive data you don't wish to share, could you please reach out to us again at the email address with the dataset and/or graph?

    The easiest way to do this is to modify the graph such that it is passing (perhaps just removing Train Anomaly Detection) - doesn't need to be valid. Then you can click PUBLISH TO GALLERY on the bottom and do so in UNLISTED mode (won't actually show up in the gallery).

    Sharing the link to this gallery entity through the amlforum@ address you've already reached out to would then allow us to copy this structure to an internal test environment.

    Please let us know if you have any concerns!

    Regards,

    AK

    Sunday, June 26, 2016 5:48 PM
    Moderator