ThingWorx Analytics Training: Module 8 part 2
- December 23, 2022
- 0 replies
- 1128 views
This video continues Module 8: Time Series Modeling of the ThingWorx Analytics Training videos. It continues to show how ThingWorx Analytics automatically transforms time series datasets into ones that are ready for machine learning. It also describes the concept of virtual sensors. It finishes by describing the time series dataset that will be used in the following modules.
This video prepares for a hands-on exercise that will be shown in a following module, In this exercise you will be shown how to train a time series model. Download and use the attached "gas_prices.csv",
This video includes chapters. Use the chapter menu in the player to jump directly to specific topics.
Chapter Summaries:
0:00 – Time Series Prediction: History Pivoting & Temporal Features
ThingWorx Analytics transforms time series data into a non-time-series format through history pivoting, incorporating past feature values, goal values, and engineered temporal metrics such as averages, derivatives, and volatility.
0:53 – Dataset Transformation: Raw vs. History-Pivoted Data
A visual comparison shows how a raw dataset containing features A and B plus goal variable G is expanded into a history-pivoted dataset that captures relationships across previous records.
1:12 – Virtual Sensors: Sensor Emulation & State Detection
Learn how virtual sensor time series models predict the value of a failed, expensive, or difficult-to-maintain sensor using data from other sensors, and how the approach supports health monitoring and state detection use cases.
2:56 – Goal History Modeling: ThingWorx Analytics 8.2+
ThingWorx Analytics 8.2 and later allows model builders to include or exclude goal variable history, enabling the current and previous goal values to be used as predictive inputs when beneficial.
3:21 – Look Back & Look Ahead: Auto Windowing and Forecast Horizons
This section explains look back and look ahead parameters, including automatic window selection using sizes of 2, 4, 8, and 16, virtual sensor scenarios, next-step forecasting, and multi-period future predictions.
5:23 – Training UI: Configuring Time Series Parameters
A walkthrough of the New Predictive Model workflow shows where to configure look back, look ahead, and goal history settings within the ThingWorx Analytics training interface.
5:48 – Static Fields: Improving Training Performance
Discover how static fields bypass history expansion and temporal feature generation, reducing unnecessary calculations and improving training speed for datasets containing attributes such as actuated mass, idle length, work cycles, and pump displacement.
7:31 – Hands-On Exercise: Predicting US East Coast Gas Prices
The exercise introduces a real-world forecasting problem using weekly US gasoline price data, with the objective of predicting East Coast gas prices one week into the future.
8:35 – Time Series CSV Structure: Entity IDs, Temporal Fields & Filters
Review the structure of the 1,026-record CSV dataset, including entity IDs, temporal fields, regional gas price columns, handling continuous versus gapped time series, and the training/scoring filter used during model validation.