Data Labeling for Machine Learning

Course details

Almost 2.5 quintillion bytes of data are produced every day—mostly raw, unlabeled data—but supervised learning techniques for machine learning require data to be labeled in order to use it for training. This makes data labeling, time-consuming and expensive though it may be, a vital part of machine learning. In this course, certified Google cloud architect and data engineer Janani Ravi guides you through how to get started with data labeling. Learn about different approaches to data labeling, as well as the challenges, best practices, and use cases that go with it. Go over data labeling with Azure ML, and find out how to set up an image labeling project and perform manual image labeling, reviews, and progress checks. Step through the full process of performing manual and ML-assisted data labeling on Azure, then explore how to use Snorkel for data labeling, including how to create diverse labeling functions and models.

Concepts

0. Introduction

01 - The need for data labeling

1. Get Started with Data Labeling

02 - The data labeling process
03 - Approaches to data labeling
04 - Data labeling challenges, best practices, and use cases
05 - Data labeling with Azure ML
06 - Setting up an Azure ML workspace
07 - Setting up an image labeling project - Creating data assets
08 - Setting up an image labeling project - Configuring settings
09 - Manual image labeling and review
10 - Manual labeling progress checks

2. Perform Manual and ML-Assisted Data Labeling on Azure

11 - Automated machine learning for image classification
12 - Examining model training metrics
13 - Data labeling project insights
14 - ML assisted labeling with clustering and pre-labeling
15 - Configuring inference for new training runs
16 - Exploring the labeled dataset

3. Use Snorkel for Data Labeling

17 - Programmatic labeling with Snorkel
18 - Installing Python libraries
19 - Exploring the spam ham dataset
20 - Writing and analyzing labeling functions
21 - Exploring other labeling functions
22 - Programmatic labeling using the majority label voter
23 - Scoring and comparing the label models

4. Create Diverse Labeling Functions and Models in Snorkel

24 - Increasing the number of labeling functions
25 - Using sentiment and parts of speech tagging in labeling functions
26 - Evaluating labeling function metrics on test data
27 - Using all labeling functions to label data
28 - Training a classifier on programmatically generated labels

Conclusion

29 - Summary and next steps

Data Labeling for Machine Learning

Authors

Janani Ravi