Generating Confidence Models via ThingWorx APIs
Overview
In the 8.5 release, ThingWorx Analytics confidence model functionality is available only via a ThingWorx API. In a future release, the functionality will be accessible in Analytics Builder as well. For now, confidence models can be created via APIs in two ways:
• Training Thing – The confidence model is generated from validation data during the same job that trains a new predictive model. The predictive model is output in PMML format and includes the confidence model information.
• Validation Thing – The confidence model is generated from an existing predictive model. A new PMML model is generated with the confidence model information included.
Create a Confidence Model
1. Log into ThingWorx Composer, click the Things icon in the left navigation panel, and open either the Training Thing or the Validation Thing.
2. Click the Services tab at the top of the screen and then open the CreateJob service.
3. Enter the following minimum parameters, based on the job type you are creating:
|
Parameter
|
Description
|
Applies to Job Type
|
||
|---|---|---|---|---|
|
jobName
|
A name to reference the job.
|
Training, Validation
|
||
|
datasetRef
|
Click to open a dialog box, select Add, and enter a datasetUri and a format (must be parquet).
|
Training, Validation
|
||
|
modelUri
|
Enter the ID of an existing predictive model. Use the following format: results:/models/<model ID>
|
Validation
|
||
|
goalField
|
Field name of the goal variable.
|
Training, Validation
|
||
|
ensembleTechnique
|
Technique for aggregating the learner output if multiple learners are used. This parameter is optional. Defaults to SOLOIST if only one learner is used. Defaults to ELITE_AVERAGE if multiple learners are used.
|
Training
|
||
|
learners
|
Click to open a dialog box, select Add, and enter at least one learningTechnique.
|
Training
|
||
|
validationHoldoutPercentage
|
Enter a number greater than 0 and less than 1 to indicate what percentage of the data will be held out for validation.
|
Training
|
||
|
createConfidenceModel
|
Click True to enable confidence model creation.
|
Training, Validation
|
||
|
confidenceLevel
|
Enter a number greater than 0 and less than 1 that will be used to calculate the likelihood of an actual value falling within a specific range. The default value is 0.8.
|
Training, Validation
|
4. Click Execute.
◦ Training Thing – A training job runs to build a predictive model and a validation job runs to generate the confidence model.
◦ Validation Thing – A validation job runs on the existing predictive model and generates a confidence model.
When the job completes, a jobID is returned in the Output panel.