AWS Certified Machine Learning Engineer MLA-C01 Notes

Domains

• Domain 1: Data Preparation for Machine Learning (ML) (28% of scored content)
• Domain 2: ML Model Development (26% of scored content)
• Domain 3: Deployment and Orchestration of ML Workflows (22% of scored content)
• Domain 4: ML Solution Monitoring, Maintenance, and Security (24% of scored content)

Services

• Amazon Data Firehose
• AWS DataSync
• AWS Glue
• Amazon SageMaker

Notes from AWS Certified Machine Learning Engineer Study Guide

Chapter 1: Introduction to Machine Learning

Knowledge

• contextuality
• applicability
• experience-based
• dynamism

ML Lifecycle

1. define ml problem
2. collect data
3. process data
4. choose algorithm
5. train model
6. evaluate model
7. deploy model
8. derive inference
9. monitor model

Structure of a Neural Network

• neuron
• input layer
• hidden layers
• output layer

Types of Neural Networks

• Artificial Neural Networks (ANN): capable of learning from data by adjusting the weights of connections between neurons to minimize error
• Deep Neural Networks: able to model more complex patterns in the data
• Convolutional Neural Networks (CNN): designed for processing grid-like data such as images
• Recurrent Neural Networks (RNN): tailored for sequential data

Chapter 2: Data Ingestion and Storage

Batch

• DataSync
• DMS

Streaming

Sources

• Clickstream data
• IOT devices
• Live gaming data

Helpful Services

• Kinesis Data Streams
• Kinesis Data Firehose
• Amazon Managed Streaming for Apache Flink

Data Quality

• Relevant
• Representative
• Rich
• Reliable
• Responsible

Ingesting and Storing Data

AWS Storage Systems

• Storage Abstractions: Data Lake, Data Lakehouse, Data Platform, Cloud Data Warehouse
• Storage Systems: HDFS, EMRFS, Object Storage, Block Storage, Streaming Storage, Cache
• Storage Low-Level Components: HDD, SSD, RAM, CPU, Networking, Compression, Serialization

Data Formats and Ingestion Techniques

Data Formats

• Semi Structured
  • CSV
  • JSON
  • JSONL
• Structured
  • Column Based
    • Apache Parquet
    • Apache Optimized Row Columnar (ORC)
  • Row Based
    • Apache Avro
    • RecordIO

Benefits of Columnar Formats

• column-specific compression
• column-specific encoding and compression
• queries only access relevant columns, improving performance

AWS DataSync Use Cases

• discover data
• migrate data
• archive cold data
• replicate data
• transfer data for timely in-cloud processing

AWS Glue Use Cases

• complex ETL pipeline development
• data discovery
• support for data processing frameworks
• simplified data engineering experience

Amazon Simple Storage Service (S3)

S3 Storage Classes

• S3 Standard: frequently accessed data, low latency, high throughput
• S3 Intelligent-Tiering: data with unknown or changing access patterns, automatic cost savings
• S3 Standard-IA (Infrequent Access): long-lived but infrequently accessed data
• S3 One Zone-IA: infrequently accessed data that does not require multiple Availability Zone
• S3 Glacier Instant Retrieval: long-term archive data that requires milliseconds retrieval
• S3 Glacier Flexible Retrieval: long-term archive data that requires minutes to hours retrieval
• S3 Glacier Deep Archive: long-term archive data that is accessed once or twice a year
• S3 Outposts: data that needs to remain on-premises for latency or data residency requirements

Amazon Athena

• serverless interactive query service
• uses standard SQL

Amazon Elastic File System (EFS)

Amazon FSx for Lustre

Amazon FSx for NetApp ONTAP

Amazon FSx for Windows File Server

Amazon FSx for OpenZFS

Amazon Elastic Block Storage (EBS)

Amazon Relational Database Service (RDS)

Amazon DynamoDB

Chapter 3: Data Transformation and Feature Engineering

Understanding Feature Engineering

• Categorical data: finite number of distinct categories
• Numerical data: anything that can be represented as a number
  • Discrete: countable values
  • Continuous: any value within a range
• Textual data: books, social media posts, articles, etc.
• Image data: pixel values
• Time series data: collection of observations or measurements recorded over regular intervals of time

Defining Features

Selecting Features for Model Training

Data Cleaning and Transformation

• Managing Missing Values
  • Collect
  • Impute
  • Drop
• Detecting and Treating Outliers
  • Delete
  • Logarithmic Transform
  • Impute
• Performing Deduplication
• Standardizing and Reformatting
  • Load Data
  • Standardize Data
  • Export Data
  • Create AWS Glue job
  • Load clean data
• Removing Noise and Errors

Feature Engineering Techniques

• Normalization
  • Equal weighting
  • Faster convergegence
  • Enhanced interoperability
• Standardization
• Scaling
  • Robust Scaling
  • MinMax Scaling
  • MaxAbs Scaling

Feature Engineering for Categorical Data

Feature Engineering for Time-Series Data

Feature Engineering for Image Data

Feature Engineering for Textual Data

• Tokenization: breaking text into smaller units (tokens)
• Stop Words Removal: eliminating common words that do not add significant meaning
• Stemming and Lemmatization: reducing words to their root form
• N-grams: contiguous sequences of n items from a given text
• Word Embeddings: representing words in a continuous vector space

Data Labeling

Amazon SageMaker Ground Truth

• Create a labeling job
• Automated Data Labeling
• Human review
• Data labeeling and validation
• Store labeled data
• Model training

Managing Class Imbalance

Class-Imbalance Mitigation Techniques

• Oversampling: increasing the number of instances in the minority class
• Undersampling: reducing the number of instances in the majority class
• Class weighting: assigning higher weights to the minority class during model training

Chapter 4: Model Selection

Understading AWS AI Services

Vision

Rekognition

Speech

Language

Chatbot

Recommendation

Personalize

Use Cases

Generative AI

Bedrock

Developing Models with Amazon SageMaker Built-in Algorithms

Supervised ML Algorithms

General Regression and Classification Algorithms

• Linear Learner
• Linear Regression
• Logistic Regression
• Support Vector Machine (SVM)

Unsupervised ML Algorithms

• Clustering
• K-Means Clustering
• Elbow Method Visualization
• Dimensionality Reduction
• Principal Component Analysis (PCA)

Criteria for Model Selection

Chapter 5: Model Training and Evaluation

Chapter 6: Model Deployment and Orchestration

Chapter 7: Model Monitoring and Cost Optimization

Chapter 8: Model Security

Services

• Database Migration Service (DMS): helps migrate databases to AWS easily and securely
• DataSync: data transfer service
• DynamoDB: NoSQL database service
• EBS: block storage service
• EFS: scalable file storage for Linux
• FSx for Lustre: high-performance file system for fast processing of large datasets
• FSx NetApp ONTAP: fully managed file storage service
• FSx for Windows File Server: fully managed Windows file system, supports SMB
• Glue DataBrew: visual data preparation tool
• Kinesis Data Streams: real-time data streaming service
• Kinesis Data Firehose: fully managed service for real-time data delivery
• Lake Formation: centralize data governance and security
• Managed Streaming for Apache Kafka (MSK): fully managed service for Apache Kafka
• Redshift: petabyte-scale data warehousing
• SageMaker Data Wrangler: simplifies data preparation and feature engineering
• SageMaker Jumpstart: pre-built templates for common ML use cases
• RDS: relational database service
• S3 Standard: object storage service
• S3 Intelligent-Tiering: automatically moves data between two access tiers when access patterns change
• S3 One Zone-IA: lower-cost option for infrequently accessed data that does not require multiple availability zone resilience

Link

Model support by AWS Region in Amazon Bedrock

Terms

• Feature: a measurable property or characteristic of the data, can be numerical or categorical
• Overfitting: model learns the training data too well, including noise and outliers, leading to poor generalization on new data
• S3 Partitioning: Apache hive-like storage patterns that may be defined in Athena
• Scaling: adjusting the range or distribution of features
• Underfitting: model is too simple to capture the underlying patterns in the data, leading to poor performance on both training and new data