none
Scalability with Sci Kit Learn and SQL 2017 RRS feed

  • Question

  • Do we have to use Microsoft provided packages such as revoscalepy to achieve high performance? Microsoft documentation says that using Microsoft packages is better because trained model code is not interpreted when called for making a prediction. We are looking for direction for using open source packages such Sci-Kit learn instead to take advantage of open source libraries.
    Friday, October 20, 2017 3:11 PM

All replies

  • By high-performance, if you are referring to "Realtime" web-services (https://docs.microsoft.com/en-us/machine-learning-server/operationalize/concept-what-are-web-services#realtime), then yes they are currently limited to microsoft packages (revoscalepy, microsoftml in Python and RevoScaleR, MicrosoftML in R) where R/Python interpreters are bypassed.

    But, for standard web-services (https://docs.microsoft.com/en-us/machine-learning-server/operationalize/concept-what-are-web-services#standard-web-services) you should see no performance differences with Microsoft packages vs. open source packages while performing predictions/scoring.

    Friday, October 20, 2017 11:17 PM