Thingworx Machine Learning r50 Release Notes

    Feature Enhancements

    Platform

    • Created a new status of Queued to represent the time a job is kicked off to the time it starts running. The Queued status will represent jobs that are submitted but have not been picked up by a worker yet. The Running status will represent jobs that are currently active.
    • The field "dataSet" has been changed from Required to Optional in the Create Filter service
    • Added the capability to upload a csv/gz file for users that have only a PostgreSQL database.
    • Added ROC curves as a comparison metric within the Prediction Model section to determine how sets of learners are compared. Available choices are now PEARSONS, RMSE, and ROC (only to be used with Boolean goals).
    • Updated error messages for file upload process to provide more detail
    • Added a new set of prediction_evaluation endpoints ( /datasets/:dataset_name/predictive_evaluation) which will run in addition to Predictive Scoring to calculate how well the scoring job ran given a filled goal column.

    SPRs

    Platform

    • Corrected an issue where some services used exclusions and other used exclusionList. All services will use Exclusions, but ExclusionList is available for backward compatibility.
    • Corrected an issue where the incorrect http status code was returned when posting duplicate data sets. Posting a duplicate dataset will now return a 400 http status message.
    • Corrected an issue where a Failed message was returned when making a status call for a resultID that does not exist.
    • Corrected an issue where Elite Average incorrectly validates training and validation datasets in prediction model generation.
    • Corrected an issue where ordinals were being denormalized twice within Profiles.
    • Corrected an issue where deleting a dataset or dataset data did not remove the content of all associated tables.