ThingWorx Machine Learning 51.0.1 Release Notes

Platform

• Added a detailed error message when a dataset or filter result fails due to no records alerting the user there are no records. The error message now shows "description": "IllegalArgumentException: DataSet has no rows. This is most likely caused by over-filtering of data. The requested filters are : [Filter[fieldName=Rx Class: MM Sympathomimetics [Y2],expression=[\"1\"],type=INCLUDE]]"

• The previous version of Profiles is now obsolete and all Profile endpoints have been updated to the enhanced version of Profiles (ProfilesV2). When users run profiles using the URI: /profiles, they will receive the results of the enhanced Profiles (previously profilesV2). This enhanced version provides more direct profiles and dynamic numeric value ranges, as well as returning profiles that are mutually exclusive.
  • Parameter treeDepth is now called maxD​epth
  • Parameter populationPercentage is now called minPopulationPercentage
  • When retrieving Profile results, variables populationPercentage, averageGoal, and treeId have been removed. The variable totalPopulationCount has been added to the results to provide users with the total number of records in a profile that match all conditions in that profile.

• Improved performance for Predictive Scoring responses by returning CSV files. JSON files will no longer be returned.
  • Retrieve Predictive Scoring Results
    • Request: URI: /datasets/:dataset_name/predictive_scores/:result_id/results; Method: GET Post body included additionalFeature: "GRINDER TYPE"
    • Response:

IDENTIFIER  DATE TIME  PREDICTED GOAL 'GRINDER ERROR OCCURRENCE' SCORE PREDICTED GOAL 'GRINDER ERROR OCCURRENCE' MIN  PREDICTED GOAL 'GRINDER ERROR OCCURRENCE' MAX  FEATURE 'GRINDER TYPE'

7946520_Jan2015 9/5/15  0.072246412 -0.002453588    0.146946412 Integrated

1547215_Jan2015 9/5/15  0.929032258 0.439332258 1.418732258 Satellite

2105350_Jan2015 9/5/15  0.072246412 -0.002453588    0.146946412 Integrated

7571180_Jan2015 9/5/15  0.091785577 0.017085577 0.166485577 Integrated

• Retrieve Predictive Evaluation Results
  • Request:URI: /datasets/:dataset_name/predictive_evaluation/:result_id/results; Method: GET Post body included additionalFeature: "GRINDER TYPE"
  • Response:

IDENTIFIER  DATE TIME  PREDICTED GOAL 'GRINDER ERROR OCCURRENCE' SCORE PREDICTED GOAL 'GRINDER ERROR OCCURRENCE' MIN  PREDICTED GOAL 'GRINDER ERROR OCCURRENCE' MAX  FEATURE 'GRINDER TYPE'

7946520_Jan2015 9/5/15  0.072246412 -0.002453588    0.146946412 Integrated

1547215_Jan2015 9/5/15  0.929032258 0.439332258 1.418732258 Satellite

2105350_Jan2015 9/5/15  0.072246412 -0.002453588    0.146946412 Integrated

7571180_Jan2015 9/5/15  0.091785577 0.017085577 0.166485577 Integrated

• Retrieve Prescriptive Scoring Results
  • Request: URI: /datasets/:dataset_name/prescriptive_scores/:result_id/results; Method: GET
  • Response:

ROW ID,GOAL NAME,ORIGINAL SCORE,OPTIMIZED SCORE,FEATURE 'REGION' ORIGINAL VALUE,FEATURE 'REGION' OPTIMAL VALUE,FEATURE 'SERVICE DESK CALLS PRIOR MONTH' ORIGINAL VALUE,FEATURE 'SERVICE DESK CALLS PRIOR MONTH' OPTIMAL VALUE,FEATURE 'SERVICE MAINTENANCE CALLS PRIOR MONTH' ORIGINAL VALUE,FEATURE 'SERVICE MAINTENANCE CALLS PRIOR MONTH' OPTIMAL VALUE

1,Grinder Error Occurrence,0.9230592998633728,0.9597735525089918,CENTRAL,WESTERN,2.0,11.999999999999998,0.0,0.0

2,Grinder Error Occurrence,5.136602782447331E-5,6.412294026640543E-5,CENTRAL,WESTERN,0.0,11.999999999999998,0.0,10.999999999999998

3,Grinder Error Occurrence,0.03129651028182871,0.255824566259157,EASTERN,WESTERN,1.0,11.999999999999998,0.0,0.0

4,Grinder Error Occurrence,0.4567072706753289,0.8318259174974278,WESTERN,WESTERN,4.0,11.999999999999998,0.0,0.0

5,Grinder Error Occurrence,0.9130724569975914,0.9546862867887936,NORTH CENTRAL,WESTERN,4.0,11.999999999999998,0.0,0.0

• Added ability to receive Important Features as a separate csv file for Predictive Scoring
  • Retrieve Predictive Scoring Important Feature Result
    • Request: URI: /datasets/:dataset_name/predictive_scores/:result_id/results_important_features; Method: GET
    • Response:

IDENTIFIER,IMPORTANT FEATURE NAME,IMPORTANT FEATURE VALUE

8635400_Jan2015,Avg Daily Cycles (Prior Week),0.46091079501598853

7946520_Jan2015,Avg Daily Cycles (Prior Week),0.4729881770353392

1057330_Jan2015,Avg Daily Cycles (Prior Week),0.31151716736922497

1852340_Jan2015,Avg Daily Cleanings (Prior Week),0.2107913897940213

• Retrieve Predictive Evaluation Important Feature Result
  • Request: URI: /datasets/:dataset_name/predictive_evaluation/:result_id/results_important_features; Method: GET
  • Response:

IDENTIFIER,IMPORTANT FEATURE NAME,IMPORTANT FEATURE VALUE

8635400_Jan2015,Avg Daily Cycles (Prior Week),0.46091079501598853

7946520_Jan2015,Avg Daily Cycles (Prior Week),0.4729881770353392

1057330_Jan2015,Avg Daily Cycles (Prior Week),0.31151716736922497

1852340_Jan2015,Avg Daily Cleanings (Prior Week),0.2107913897940213

• Added a more descriptive error message for unsupported file types during upload. The error message has been updated to state "The file type selected is not supported, please use txt, csv, gzip."

• Added the ability for users to kill either a single job or all jobs in the queue for that appId and appKey. Terminated jobs will report a status of TERMINATED.

Extensions

• Updated ThingWorx Machine Learning Extensions to include ComparisonMetric as an input parameter and output values for Prediction Model Generation

Platform

• Developer Edition: Corrected an issue where concurrently run jobs were reporting errors but still completing successfully.
• Resolved an issue where PVA results were incorrect on un-optimized datasets.

Backward Compatibility

• The original version of Profiles has been removed and all endpoints have been updated to use the enhanced capabilities of the new version of Profiles. The endpoints, using the enhanced version, are /profiles.
• Predictive Scoring, Predictive Evaluation, and Prescriptive Scoring will now return a csv file instead of a json file to improve performance for large sets of data. JSON files will not be returned. Any previously run scoring jobs will no longer be accessible and will need to be re-run.