Coursera

Week 1

Popular datasets

TFDS (TensorFlow Datasets):

• Ready to use
• Flexible
• Standardized input pipelines
• Plethora of public research data
• Seamless integration
• Faster prototyping

Data pipelines

ETL for TensorFlow:

• Extract
• Transform
• Load

Versioning

mnist = tfds.load("mnist:1.*.*")

• 1: Major version
• First asterisk: Minor version
• Second asterisk: patch version

Format data to data and label:

dataset = tfds.load("mnist", as_supervised=True)

As a dictionary:

dataset = tfds.load("mnist", as_supervised=False)

DatasetBuilder

Pick dataset -> Download -> Extract dataset