ThingWorx Analytics Training: Module 4 Part 1
- December 23, 2022
- 0 replies
- 1067 views
This video begins Module 4: Data Transformation & Feature Engineering of the ThingWorx Analytics Training videos. It describes what data transformation is, and how feature engineering can improve machine learning models. You will learn about independent and dependent variables in your data, and how an "analytics ready view" looks for use with ThingWorx Analytics.
This video contains a hands-on exercise, and it is recommended to complete it before moving on to subsequent sections. This exercise involves taking an analytics ready view in .csv format, and using it to create a ThingWorx Analytics dataset. Download and use the attached "BEAN_PRO_DATA_PROCESSED.csv", or use the file you modified in Module 3 Part 3.
This video has chapters — open the chapter menu in the player to jump to any section.
Chapter Summaries
0:00 – Module 4 Intro: Data Transformation & Feature Engineering Overview
Introduces Module Four and its goals: defining data transformation and feature engineering for machine learning models, plus independent variables, dependent variables, topics, and the analytics-ready view used by ThingWorx Analytics.
0:32 – Feature Engineering Basics: Building Features from the Contextual Repository
Explains how feature engineering pulls data from the contextual repository to build new columns or features—static or time series, like service records—that help ThingWorx Analytics identify the factors driving the goal, including manual processing done outside the tool.
1:08 – ThingWorx Analytics Data Flow: Signals, Profiles & Champion Model
Walks through the typical data flow from raw source data to a structured analytics-ready view combined with metadata, then using signals, profiles, and predictive modeling to iterate toward a best-performing champion model embedded in a smart connected application.
1:55 – Topics & Goal Variables: Defining the Outcome to Predict
Defines a topic as the outcome or goal variable you want to predict, with examples such as forecasting machine component failure or whether a patient will require emergency medical treatment.
2:18 – Independent vs. Dependent Variables: Features & Data Types
Distinguishes independent variables (features) from the dependent goal variable, using the calorie-intake versus body-weight example, and covers data types—strings, numbers, Booleans—noting strings can't be predicted mathematically as a goal.
3:11 – Analytics-Ready View & JSON Metadata: Python/R-Style Data Prep
Describes building a clean analytics-ready dataset with an accompanying JSON metadata file, mirroring workflows in Python or R—profiling data, transforming and aggregating features, applying business rules—and points back to Module Three for metadata parameters.
4:05 – BeanPro Use Case: Predicting Grinder, Pump & Heating Element Failure
Introduces the Module Two BeanPro espresso machine use case, with grinder failure as the dependent variable surrounded by independent variables, aiming to predict failure likelihood in the grinder, pump, or heating elements.
4:45 – Analytics Workflow: Use Case, Goal Variable & Historical Data
Outlines identifying a use case before feature engineering—assessing business value, impact, and effort—then understanding the data and business rules for the goal variable and supplying historical results needed to train the model.
5:23 – Aggregation Level: Actionable Windows & Sensor Data History
Covers choosing an aggregation level tied to an actionable prediction window, aggregating granular sensor data from connected devices to match that interval, and including enough historical examples of the predicted condition to train a model.
6:17 – Independent Variables & Levers: Descriptive/Prescriptive Attributes & Prescriptive Scoring
Explains deriving features through binning and pivoting, adding descriptive attributes like hardware details, and prescriptive attributes called levers—such as BeanPro weekly usage—that prescriptive scoring can adjust to an optimal value to impact the goal.
7:33 – Pivoting: One-to-Many Service Records & Aggregate Features
Details pivoting to resolve one-to-many relationships between machine and service tables, engineering new aggregate features like the count or average of times a part was replaced, performed before loading data into ThingWorx Analytics.
8:09 – Binning: Continuous-to-Categorical Buckets & Noise Reduction
Shows how binning transforms continuous values into categorical groups or buckets—like machine age ranges of 0–5 and 5–10 years—then applies an aggregate such as count or average to reduce noise and surface important patterns.
8:51 – Metadata & Sampling: Detect Metadata Service, Identifier Fields & Analytics-Ready View
Reviews supplying metadata via a manual JSON file, Analytics Builder, or the Detect Metadata service in the Data Thing, recommending an identifier field and opt types for time series data, plus sampling large datasets and reading a sample analytics-ready view.
9:57 – Loading BeanPro CSV in Analytics Builder: Metadata Inference & Opt Types
Demonstrates loading the BeanPro CSV in Analytics Builder—Data > New, naming the dataset, Choose File, leaving Upload Metadata unchecked to infer metadata, flagging time series data, submitting, then reviewing inferred Opt Type and Data Type with Restore Defaults and JSON download.