ThingWorx Analytics Training: Module 6 Part 5

Chapter Summaries

0:01 – Create ThingWorx Analytics Dataset: Iris CSV Upload
Walks through the Analytics > Data area to create a new ThingWorx Analytics dataset named "Iris Data," uploading the Iris CSV while unchecking the upload-metadata and time-series options before submitting.

0:51 – Reviewing Inferred Metadata & Dataset Details
Shows how ThingWorx Analytics auto-infers metadata — sepal/petal length and width as DOUBLE/continuous and class as STRING/categorical — then confirms the dataset URI, row count, field count, and the time-series Boolean flag.

2:00 – Training a Categorical Model via Microservices: Create Job
Explains that categorical models can't be trained in the Builder UI and must use microservices, navigating Browse > Things > Analytics Server Training Thing > Services to execute the Create Job service (job name Iris_NN).

3:01 – Dataset Ref, PARQUET Format & Goal Field
Adds the Data Set Ref by pasting the copied dataset URI, setting the format to lowercase "parquet," warning about leading/trailing spaces, skipping filters and exclusions, and entering "class" as the goal field.

4:15 – Ensemble Technique SOLOiST & Neural Net Learner
Sets the Ensemble Technique to SOLOiST and adds a learner using the NEURAL_NET learning technique, noting that other neural net parameters are left at ThingWorx Analytics default values.

5:17 – Model Parameters: Max Allowed Fields & Time Series Settings
Covers leaving Comparison Metric blank for a soloist model, keeping Lookback/Lookahead and Virtual Sensor unchanged since it isn't a time-series model, and explaining Max Allowed Fields (25 default) with all four Iris independent variables and the redundancy filter.

6:27 – Validation Holdout (80/20) & Confusion Matrix Weights
Sets the validation holdout to 0.2 for the common 80/20 split, notes that sampling and anomaly-detection params (Thing Watcher) don't apply, leaves the create-confidence-model option false, and reviews the four equally weighted confusion matrix quadrants.

7:52 – Saving the Input Set & Executing the Training Job
Sets validate-on-training to false to get validation-subset metrics, saves parameters as Iris_NN_Parameters, executes the job, then copies the model job ID and uses Get Job Infos to capture the training and validation job IDs.

9:22 – Retrieving Validation Metrics: MCC & Accuracy
Opens the Validation Thing, runs the Retrieve Result service with the validation job ID, and reviews goal Class across 33 records (~20%), including the Matthews correlation coefficient and model accuracy in the validation metrics info table.

10:43 – Confusion Matrix & Model Evaluation
Opens the confusion matrix and labels info tables to reveal the three iris classes in a 3x3 matrix, showing a single misclassification (versicolor predicted for a virginica) with the strong diagonal confirming a very accurate model.

11:50 – Batch Predictive Scoring Job Setup
Starts a batch predictive scoring job on the Analytics Server Prediction Thing via the Batch Score service (job name Iris_Scores), adding the dataset ref with the pasted URI and lowercase "parquet" format with no filter.

13:23 – Model URI, Input Set & Executing the Scoring Job
Enters goal field "class," builds the Model URI as results:/models/ plus the copied model URI, keeps important field count at zero (unavailable for categorical models), saves the Iris_Scores_Input set, executes, and captures the scoring job ID.

15:23 – Interpreting Prediction Scores by Iris Species
Runs Retrieve Result to display per-record scores, showing the numeric risk that each flower belongs to setosa, versicolor, or virginica — the first record scoring a very high probability of setosa — concluding the exercise.

15:49 – Module Wrap-Up: Predictive Models & Model Validation
Closes out Module Six of the ThingWorx Analytics training on predictive models and model validation, previews upcoming modules, and points viewers to a PTC customer success representative for more information.