Principal Component Analysis (PCA)

Introduction to Principal Component Analysis (PCA) (Lesson 129)

• PCA transforms a dataset with many features into a dataset with fewer, new features.
• PCA components are less interpretable compared to original features.
• Only works with numeric continuous data, not suitable for categorical values.
• Data should be normalized for PCA to be effective.

SageMaker PCA Details

• Modes in SageMaker:
• Regular Mode: For sparse and moderate-sized datasets.
• Random Mode: For very large datasets, using an approximation algorithm.
• Data Formats:
• Training Data: CSV and protobuf RecordIO.
• Inference Data: CSV, JSON, protobuf RecordIO.