ThingWorx Analytics Training: Module 8 Part 3

Chapter Summaries

0:00 – Getting Started: Navigate to the ThingWorx Analytics Data Section
Kicks off the walkthrough by opening ThingWorx Analytics and using the left-side Analytics icon to reach the Data area where datasets are created and managed.

0:16 – Create a Dataset: Upload US Gas Prices CSV & Enable Time Series Data
Shows how to create a new dataset (named "US Gas Prices"), attach the source CSV with Choose File, uncheck the upload-metadata option, tick the Time Series Data box, and Submit so ThingWorx Analytics infers the metadata.

0:55 – Configure Metadata: Op Types, Entity ID, Temporal & Static Fields
Walks through refining the inferred metadata—switching entity to Entity ID, setting the week field to a temporal op type with a sampling interval of one, marking Record Purpose as informational, and explaining when to flag fields as static.

2:39 – Save Metadata & Create Data Set: Download as JSON and Review the Dataset URI
Covers Restore Defaults for fixing mistakes, using Download as JSON to reproduce the dataset in another environment alongside the CSV, then clicking Create Data Set and reviewing its URI, row and field counts, and time-series Boolean flag.

3:34 – Create a Model: Neural Network Learner, Dataset & Goal Variable Selection
Demonstrates building a new model (US Gas Prices_NN) with a neural network learner, selecting the newly created dataset from the dropdown, and choosing the US East Coast region as the continuous goal variable.

4:36 – Build a Training Filter: Record Purpose, Include & Exclude Records
Explains creating a Training_Data filter from the Record Purpose field—choosing the training value with Include Records—and notes that selecting Scoring with Exclude Records produces an equivalent filter.

5:51 – Exclude Fields & Prevent Data Leakage in Time Series Models
Confirms the time-series configuration, uses Exclude Fields to drop Record Purpose from training, and explains why leakage-prone fields (like data appended after asset failure) must be excluded to avoid unrealistically high performance metrics.

7:28 – Advanced Model Configuration: Validation Holdout, Max Fields & Redundancy Filter
Reviews the Advanced Model Configuration tab—the 20% validation holdout from the 80/20 split, the Max Fields signals cap (changed to six for practice), and the optional Redundancy Filter for selecting top mutually-informative signals.

9:07 – Confidence Models, Sampling & Time Series Parameters (Lookback/Lookahead)
Notes that confidence models, sampling strategy, and confusion matrix weights apply only to Boolean goals (grayed out for this continuous goal), then sets time-series parameters—lookback of four weeks, lookahead of one—with Use Goal History enabled.

10:27 – Learning Techniques: Add Learner, Random Forest, Ensemble & Soloist Neural Network
Explores the learning techniques section via Add Learner, showing random forest options (max depth, number of trees), the elite average/average/best ensemble methods, and removing all learners except the neural network to create a soloist model.

12:05 – Train & Review the Model: Submit, RMSE Results & Randomized Weights
Submits the model for training and reviews the completed output—goal variable, filter, validation record count, creation date, and an RMSE of 0.0496 (about five cents off), explaining why randomized neural-network weights yield slightly different metrics each run.

13:41 – Chronological Train/Validation Split for Time Series Data
Explains that ThingWorx Analytics splits time-series datasets chronologically—first 80% of records for training and the final 20% (196 records here) for validation—unlike the splitting used for non-time-series models.

14:09 – Performance Metrics: Validation vs Training, Bubble/Pie Charts & Overfitting
Opens the model results to compare the Validation Metrics and Training Metrics tabs, reads the actual-vs-predicted bubble chart and the ±25% accuracy pie chart, and warns that a large gap between the two RMSE values signals overfitting rather than generalization.

16:38 – Wrap-Up: Module 8 Time Series Complete & Next Steps
Closes the exercise, confirms completion of Module 8 (Time Series) of the ThingWorx Analytics training, previews upcoming course modules, and points viewers to a PTC customer success representative for more information.