AWS Certified Machine Learning - Specialty (MLS-C01) - Fact Sheet¶

⚠️ RETIRED April 15, 2025. No longer available for new candidates. Replaced by AWS Machine Learning Engineer - Associate (MLA-C01). Material preserved as historical reference.

Quick Reference¶

Exam Details¶

Exam Code: MLS-C01
Duration: 180 minutes (3 hours)
Number of Questions: 65 questions
Passing Score: 750/1000
Question Format: Multiple choice, multiple response
Cost: $300 USD
Validity: 3 years
Prerequisites: Recommended 1-2 years hands-on ML/DL on AWS

Exam Domains¶

Domain	% of Exam
Domain 1: Data Engineering	20%
Domain 2: Exploratory Data Analysis	24%
Domain 3: Modeling	36%
Domain 4: Machine Learning Implementation & Operations	20%

Domain 1: Data Engineering (20%)¶

Data Repositories for ML¶

Amazon S3¶

Primary data lake - Store raw, processed, and model artifacts
📖 S3 for ML Data - Best practices for ML datasets
S3 Select - Query CSV/JSON/Parquet without full download
📖 S3 Select Documentation
Lifecycle policies - Archive old training data to Glacier
Versioning - Track dataset versions over time

Amazon RDS / Aurora¶

Structured ML features - Store feature stores and metadata
📖 Aurora ML - Invoke SageMaker directly from SQL
Aurora Serverless - Auto-scaling for variable ML workloads

Amazon Redshift¶

Data warehouse - Aggregate features from multiple sources
📖 Redshift ML - CREATE MODEL SQL syntax
Redshift Spectrum - Query S3 data lakes directly

Amazon DynamoDB¶

Real-time feature store - Low-latency feature retrieval
📖 DynamoDB for ML - Online feature store patterns
TTL - Automatically expire old features

Amazon Timestream¶

Time-series data - IoT sensor data, application metrics
📖 Timestream Documentation

Data Ingestion and Transformation¶

AWS Glue¶

ETL service - Serverless data preparation
📖 Glue for ML - ETL job authoring
Glue DataBrew - Visual data preparation
📖 DataBrew Transformations - 250+ pre-built transformations
Glue Crawler - Auto-discover schema
Glue Data Catalog - Centralized metadata repository

Amazon EMR¶

Big data processing - Spark, Hadoop, Presto
📖 EMR for ML - Spark MLlib on EMR
EMR Notebooks - Jupyter-based data exploration

AWS Data Pipeline¶

Workflow orchestration - Schedule and monitor pipelines
📖 Data Pipeline Documentation

Amazon Kinesis¶

Streaming data - Real-time ML inference pipelines
📖 Kinesis Data Streams - Real-time data ingestion
📖 Kinesis Data Firehose - Load streams to S3/Redshift
Kinesis Data Analytics - SQL or Apache Flink for streaming ETL

AWS Lambda¶

Serverless transforms - Lightweight data preprocessing
📖 Lambda for ML - Trigger on S3 events

Data Formats and Optimization¶

File Formats¶

Parquet - Columnar, compressed, best for analytics (10x faster than CSV)
📖 Parquet Format - Column pruning and predicate pushdown
ORC - Optimized Row Columnar, Hive-optimized
Avro - Row-based, schema evolution
RecordIO-Protobuf - SageMaker pipe mode format
📖 SageMaker File Formats - Recommended formats

Data Partitioning¶

Hive-style partitioning - s3://bucket/year=2023/month=01/data.parquet
📖 Athena Partitioning - Reduce scan costs

Data Security and Compliance¶

Encryption¶

S3 encryption - SSE-S3, SSE-KMS, SSE-C
📖 S3 Encryption
SageMaker encryption - Encrypt training data, models, endpoints
📖 SageMaker Encryption

Access Control¶

IAM policies - Least privilege access
S3 bucket policies - Resource-based access control
VPC endpoints - Private connectivity without internet
📖 SageMaker VPC - Deploy in private subnets

Data Privacy¶

AWS Macie - Discover and protect PII
📖 Macie Documentation
Anonymization - Remove/hash PII before training

Domain 2: Exploratory Data Analysis (24%)¶

Data Visualization and Analysis¶

Amazon SageMaker Data Wrangler¶

Visual data prep - 300+ built-in transformations
📖 Data Wrangler Guide - Import, transform, analyze
📖 Data Wrangler Transformations - Encode, scale, handle missing
Quick Model - Train XGBoost to assess feature importance
Data Quality Report - Detect anomalies, outliers, duplicates
Target leakage detection - Identify features too correlated with target

Amazon SageMaker Canvas¶

No-code ML - Business analysts build models
📖 Canvas Documentation

Amazon QuickSight¶

BI dashboards - Visualize model performance metrics
📖 QuickSight ML Insights - Anomaly detection, forecasting
📖 Q (QuickSight NLP) - Natural language queries

Amazon Athena¶

Serverless SQL - Query S3 data lakes
📖 Athena ML - USING FUNCTION for SageMaker inference

Jupyter Notebooks¶

SageMaker Studio - Integrated Jupyter environment
📖 SageMaker Studio - Code, debug, visualize
SageMaker Notebook Instances - Managed Jupyter on EC2

Statistical Analysis¶

Descriptive Statistics¶

Central tendency - Mean, median, mode
Dispersion - Variance, standard deviation, IQR
Distribution - Skewness, kurtosis
Correlation - Pearson, Spearman correlation coefficients
📖 Pandas Profiling - Automated EDA reports

Data Quality Issues¶

Missing values - Imputation strategies (mean, median, mode, forward-fill)
📖 Handling Missing Data - Scikit-learn imputers
Outliers - Z-score, IQR method, isolation forest
Imbalanced classes - SMOTE, undersampling, class weights
📖 Imbalanced-learn - Resampling techniques
Duplicates - Detect and remove duplicate records
Data drift - Monitor feature distributions over time

Feature Engineering¶

Numerical transformations - Log, square root, polynomial features
Scaling - StandardScaler, MinMaxScaler, RobustScaler
Encoding categorical - One-hot, ordinal, target encoding
📖 Feature Engineering - Data Wrangler techniques
Binning - Convert continuous to categorical
Feature crosses - Interaction features (A × B)
Time features - Extract hour, day, month, season

Dimensionality Reduction¶

PCA (Principal Component Analysis) - Linear dimensionality reduction
📖 SageMaker PCA - Built-in algorithm
t-SNE - Non-linear, visualization in 2D/3D
Feature selection - Remove low-variance, high-correlation features
📖 Feature Selection Methods

Data Splitting Strategies¶

Train/Validation/Test Split¶

70/15/15 or 80/10/10 - Typical splits
Stratified split - Preserve class distribution
Time-based split - For time-series (no random shuffle)
📖 Scikit-learn Split

Cross-Validation¶

K-fold CV - Split into k folds, train on k-1
Stratified K-fold - Maintain class balance
Time-series CV - Walk-forward validation
Leave-one-out CV - For small datasets

Domain 3: Modeling (36%)¶

Amazon SageMaker Built-in Algorithms¶

Linear Models¶

Linear Learner - Linear regression, logistic regression, multiclass
📖 Linear Learner - L1/L2 regularization, auto-tuning
Use cases: Regression, binary/multiclass classification

Tree-Based Models¶

XGBoost - Gradient boosting, best for tabular data
📖 XGBoost Algorithm - Hyperparameters, tuning
Hyperparameters: max_depth, eta (learning rate), subsample, num_round
📖 XGBoost Tuning

Deep Learning¶

Image Classification - ResNet, transfer learning
📖 Image Classification - Built-in CNN
Object Detection - Single Shot Detector (SSD)
📖 Object Detection - Bounding boxes
Semantic Segmentation - Pixel-level classification
📖 Semantic Segmentation
Seq2Seq - Machine translation, text summarization
📖 Sequence to Sequence - RNN/LSTM/GRU

Clustering and Anomaly Detection¶

K-Means - Unsupervised clustering
📖 K-Means Algorithm - Web-scale clustering
Random Cut Forest (RCF) - Anomaly detection
📖 Random Cut Forest - Time-series anomalies
IP Insights - Detect anomalous IP usage patterns
📖 IP Insights

NLP Algorithms¶

BlazingText - Word2Vec, text classification
📖 BlazingText - FastText implementation
Modes: Word2Vec (unsupervised), text classification (supervised)
Object2Vec - General-purpose embeddings
📖 Object2Vec - Sentence similarity, recommendations

Recommendation Systems¶

Factorization Machines - Sparse data, click prediction
📖 Factorization Machines - High-dimensional sparse datasets
Neural Topic Model (NTM) - Topic modeling
📖 NTM Algorithm - Document classification

Time-Series Forecasting¶

DeepAR - Probabilistic forecasting with RNNs
📖 DeepAR Forecasting - Multiple time-series
Produces: Quantile predictions (P10, P50, P90)

Deep Learning Frameworks¶

TensorFlow¶

📖 TensorFlow on SageMaker - Bring your own TensorFlow script
Keras API - High-level API on TensorFlow
TensorFlow Serving - Production model serving

PyTorch¶

📖 PyTorch on SageMaker - Custom PyTorch training
Distributed training - Data parallel, model parallel
📖 Distributed PyTorch

Apache MXNet¶

📖 MXNet on SageMaker
Gluon API - Imperative and symbolic programming

Hugging Face¶

📖 Hugging Face on SageMaker - Pre-trained NLP models
Transformers - BERT, GPT, T5, RoBERTa
Use cases: Text classification, NER, Q&A, summarization

Model Training Strategies¶

Training Job Types¶

SageMaker Training Jobs - Fully managed training
📖 Training Jobs - Automatic provisioning
Spot Instances - Up to 90% cost savings with managed spot training
📖 Managed Spot Training

Distributed Training¶

Data parallelism - Replicate model, split data across GPUs
📖 Data Parallel Library - Near-linear scaling
Model parallelism - Split large models across GPUs
📖 Model Parallel Library - For models too large for 1 GPU
Parameter servers - Asynchronous SGD with parameter servers

Training Input Modes¶

File mode - Download full dataset to training instance
Pipe mode - Stream data from S3 (faster, less disk usage)
📖 Pipe Mode - RecordIO-Protobuf format
Fast File mode - S3 FUSE mount (random access)

Hyperparameter Optimization¶

SageMaker Automatic Model Tuning - Bayesian optimization
📖 Hyperparameter Tuning - Find best hyperparameters
📖 Tuning Strategies - Bayesian, random, grid search
Hyperband - Early stopping for low-performing trials
Warm start - Use previous tuning job results
Objective metrics - Maximize accuracy, minimize loss

Model Evaluation Metrics¶

Classification Metrics¶

Accuracy - (TP + TN) / Total
Precision - TP / (TP + FP) - "Of predicted positive, how many correct?"
Recall - TP / (TP + FN) - "Of actual positive, how many found?"
F1 Score - 2 × (Precision × Recall) / (Precision + Recall)
📖 Classification Metrics
ROC-AUC - Area under ROC curve (TPR vs FPR)
PR-AUC - Area under precision-recall curve (better for imbalanced)
Confusion matrix - TP, TN, FP, FN breakdown

Regression Metrics¶

MAE (Mean Absolute Error) - Average absolute difference
MSE (Mean Squared Error) - Average squared difference
RMSE (Root Mean Squared Error) - Square root of MSE
R² (R-squared) - Proportion of variance explained
📖 Regression Metrics

Ranking Metrics¶

MAP (Mean Average Precision) - Quality of ranked results
NDCG (Normalized Discounted Cumulative Gain) - Ranking quality with position weighting

Regularization and Overfitting¶

Techniques to Prevent Overfitting¶

L1 regularization (Lasso) - Adds sum of absolute weights to loss
L2 regularization (Ridge) - Adds sum of squared weights to loss
Elastic Net - Combines L1 + L2
Dropout - Randomly drop neurons during training (neural networks)
Early stopping - Stop when validation loss stops improving
Data augmentation - Synthetic data variations (images, text)
Cross-validation - Robust performance estimation

Transfer Learning and Pre-trained Models¶

Transfer Learning¶

Use pre-trained models - ImageNet for computer vision, BERT for NLP
📖 Transfer Learning Guide
Fine-tuning - Retrain last layers on your dataset
Feature extraction - Use pre-trained as feature extractor

AWS AI Services (Pre-trained Models)¶

Amazon Rekognition - Image and video analysis
📖 Rekognition - Face detection, object detection, content moderation
Amazon Comprehend - NLP (sentiment, entities, key phrases)
📖 Comprehend - Custom classification and entity recognition
Amazon Translate - Neural machine translation
Amazon Polly - Text-to-speech
Amazon Transcribe - Speech-to-text
📖 Transcribe - Custom vocabulary, real-time
Amazon Textract - OCR and form extraction
📖 Textract - Tables, forms, key-value pairs
Amazon Forecast - Time-series forecasting
📖 Forecast - AutoML for forecasting
Amazon Personalize - Real-time recommendations
📖 Personalize - User-item interactions
Amazon Lex - Conversational AI chatbots
Amazon Kendra - Intelligent search service

Domain 4: ML Implementation & Operations (20%)¶

Model Deployment¶

SageMaker Hosting Options¶

Real-time endpoints - Low-latency predictions
📖 Real-time Inference - Deploy model behind HTTPS
Serverless inference - Auto-scaling, pay-per-request
📖 Serverless Inference - No instances to manage
Batch transform - Process large datasets asynchronously
📖 Batch Transform - No persistent endpoint
Asynchronous inference - Queue requests, long processing times
📖 Async Inference - Up to 1 hour timeout

Endpoint Configuration¶

Multi-model endpoints - Host multiple models on one endpoint (cost savings)
📖 Multi-Model Endpoints - Dynamic model loading
Multi-container endpoints - Serial or direct inference pipeline
📖 Multi-Container Endpoints
Auto-scaling - Scale based on invocations or custom metrics
📖 Endpoint Auto Scaling
Instance types - ml.t2.medium (dev) to ml.p4d.24xlarge (GPU inference)

Deployment Strategies¶

Blue/green deployment - New endpoint, shift traffic gradually
📖 Blue/Green Deployment - Canary, linear, all-at-once
Shadow testing - Send production traffic to new model without affecting users
A/B testing - Split traffic between model variants
📖 A/B Testing - Production variants

Model Monitoring¶

Amazon SageMaker Model Monitor¶

Data quality monitoring - Detect data drift
📖 Data Quality Monitor - Baseline vs production data
Model quality monitoring - Detect model performance degradation
📖 Model Quality Monitor - Ground truth labels
Bias drift monitoring - Detect bias in predictions
Feature attribution drift - SHAP value changes
📖 Model Monitor - Continuous monitoring

Amazon CloudWatch¶

Metrics - Invocations, latency, errors
📖 SageMaker CloudWatch Metrics
Alarms - Trigger notifications on thresholds
Logs - Training logs, endpoint logs

MLOps and CI/CD¶

SageMaker Pipelines¶

ML workflows - Orchestrate data prep, training, evaluation, deployment
📖 SageMaker Pipelines - CI/CD for ML
Pipeline steps - Processing, training, tuning, model, condition, callback
📖 Pipeline Steps
Model Registry - Versioning, approval workflow
📖 Model Registry - Approved, pending, rejected

SageMaker Projects¶

MLOps templates - Pre-built CI/CD with CodePipeline, CodeBuild
📖 SageMaker Projects - Model building, deployment

Model Versioning¶

Model Registry - Track model lineage
Model packages - Bundle model artifacts, inference code
Approval status - Require approval before production deployment

Edge Deployment¶

SageMaker Edge Manager¶

Deploy to edge devices - IoT, mobile, on-premises
📖 Edge Manager - Manage models on fleets
Edge Agent - Runtime for model inference on device
Model optimization - Compile with SageMaker Neo

SageMaker Neo¶

Model compilation - Optimize for target hardware
📖 SageMaker Neo - Up to 2x performance
Supported frameworks - TensorFlow, PyTorch, MXNet, XGBoost
Supported devices - ARM, Intel, Nvidia, Xilinx, edge devices

AWS IoT Greengrass¶

Edge runtime - Deploy Lambda functions and models to edge
📖 Greengrass ML Inference - Local inference

Security and Compliance¶

IAM and Access Control¶

IAM roles - Execution role for training jobs, hosting
📖 SageMaker IAM Roles
Resource policies - Control access to models, endpoints
Condition keys - Fine-grained access control

Network Isolation¶

VPC mode - Training and hosting in private subnets
📖 VPC Training and Hosting - No internet access
VPC endpoints - Private connectivity to S3, ECR
Network isolation - Block all network traffic from containers

Data Protection¶

Encryption at rest - S3, EBS volumes encrypted with KMS
Encryption in transit - TLS 1.2 for all API calls
📖 SageMaker Security

Compliance¶

HIPAA eligible - Healthcare workloads
PCI DSS - Payment card data
SOC, ISO, FedRAMP - Various compliance programs
📖 SageMaker Compliance

Cost Optimization¶

Training Cost Optimization¶

Managed spot training - Up to 90% savings
Checkpointing - Resume from checkpoint if spot interrupted
📖 Spot Training Best Practices
Rightsizing instances - Start small, scale up if needed

Inference Cost Optimization¶

Multi-model endpoints - Share infrastructure across models
Serverless inference - Pay only for compute time used
Auto-scaling - Scale down during low traffic
Batch transform - No persistent endpoint costs

Storage Cost Optimization¶

S3 lifecycle policies - Archive old data to Glacier
Delete intermediate data - Training outputs, checkpoints

Common Exam Scenarios¶

Scenario 1: Data Preparation Pipeline¶

Problem: Large dataset in S3, needs ETL before training Solution: - Use AWS Glue for serverless ETL - Glue Crawler discovers schema, populates Data Catalog - Glue job transforms data to Parquet with Hive partitioning - Store in S3, query with Athena for validation - 📖 Glue ETL Best Practices

Scenario 2: Hyperparameter Tuning at Scale¶

Problem: Need to find best hyperparameters for XGBoost Solution: - SageMaker Automatic Model Tuning with Bayesian optimization - Define hyperparameter ranges (max_depth: 3-10, eta: 0.01-0.3) - Specify objective metric (validation:auc) - Use managed spot training for cost savings - 📖 XGBoost Tuning Guide

Scenario 3: Real-time Image Classification¶

Problem: Deploy image classification model for low-latency inference Solution: - Use SageMaker built-in Image Classification or custom CNN - Transfer learning from ResNet-50 pre-trained on ImageNet - Deploy to real-time endpoint with GPU instance (ml.p3.2xlarge) - Enable auto-scaling based on InvocationsPerInstance - A/B testing with production variants - 📖 Image Classification Deployment

Scenario 4: Batch Predictions on Large Dataset¶

Problem: Score 1 million records without persistent endpoint Solution: - Use SageMaker Batch Transform - Input data in S3 (CSV or JSON Lines format) - Specify instance type and count for parallelization - Output predictions to S3 - No endpoint costs after job completes - 📖 Batch Transform Best Practices

Scenario 5: Model Monitoring and Retraining¶

Problem: Detect model drift and trigger retraining Solution: - SageMaker Model Monitor for data quality monitoring - Set baseline from training data - Schedule monitoring jobs (hourly, daily) - CloudWatch alarm triggers when drift detected - Lambda function triggers SageMaker training job - SageMaker Pipelines for automated retraining workflow - 📖 Model Monitor Drift Detection

Scenario 6: Multi-Class Imbalanced Classification¶

Problem: Dataset has 1000:1 class imbalance Solution: - SMOTE (Synthetic Minority Over-sampling) or undersampling - Use class weights in loss function - Evaluation metrics: F1 score, PR-AUC (not accuracy!) - Consider anomaly detection (Random Cut Forest) if extreme imbalance - 📖 Handling Imbalanced Data

Scenario 7: Distributed Training for Large Model¶

Problem: Train 10B parameter Transformer model Solution: - Use SageMaker model parallel library - Pipeline parallelism or tensor parallelism - Multi-GPU instances (ml.p4d.24xlarge with 8x A100 GPUs) - Distributed data parallel for data parallelism across nodes - 📖 Model Parallel Training

Scenario 8: Secure ML in VPC¶

Problem: Healthcare data, must stay in private network Solution: - VPC mode for training and hosting - Private subnets with no internet gateway - VPC endpoints for S3, SageMaker, ECR - KMS encryption for S3, EBS volumes - IAM roles with least privilege - 📖 VPC Security Configuration

Exam Tips¶

Key Topics to Master¶

SageMaker algorithms - When to use each (XGBoost, Linear Learner, DeepAR, etc.)
Data formats - Parquet vs CSV, RecordIO-Protobuf for pipe mode
Hyperparameter tuning - Bayesian vs random, objective metrics
Evaluation metrics - Precision vs recall, when to use F1 vs AUC
Deployment options - Real-time vs batch vs serverless vs async
Model monitoring - Data drift, model drift detection
Security - VPC, encryption, IAM roles
Cost optimization - Spot training, serverless inference, multi-model endpoints

Common Pitfalls¶

Confusing accuracy with F1 - Use F1 or PR-AUC for imbalanced classes
Not using spot training - Always consider for cost savings
Wrong deployment type - Batch transform for large one-time scoring, not real-time
Forgetting data drift - Production models need continuous monitoring
Not using pipe mode - Faster training for large datasets vs file mode

Calculation and Formula Questions¶

Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
F1 Score = 2 × (Precision × Recall) / (Precision + Recall)
Training time estimate - Consider instance type, data size, algorithm complexity
Cost estimate - Instance hours × instance price (don't forget spot savings)

AWS AI Services Decision Tree¶

Pre-trained model sufficient? → Use AWS AI service (Rekognition, Comprehend, etc.)
Need custom model on tabular data? → SageMaker XGBoost or Linear Learner
Need custom deep learning? → SageMaker with TensorFlow/PyTorch
Need transfer learning? → SageMaker built-in (Image Classification, BlazingText)
Need AutoML? → SageMaker Autopilot or Canvas

Algorithm Selection Guide¶

Problem Type	Best Algorithm(s)	Notes
Binary classification (tabular)	XGBoost, Linear Learner	XGBoost usually best
Multi-class classification (tabular)	XGBoost, Linear Learner
Regression	XGBoost, Linear Learner
Image classification	Image Classification (ResNet)	Transfer learning from ImageNet
Object detection	Object Detection (SSD)	Bounding boxes
Text classification	BlazingText	Fast, scalable
Sentiment analysis	BlazingText, Comprehend	Comprehend for pre-trained
Time-series forecasting	DeepAR, Forecast	Forecast for AutoML
Anomaly detection	Random Cut Forest, IP Insights	RCF for time-series
Clustering	K-Means	Web-scale
Recommendations	Factorization Machines, Personalize	Personalize for pre-built
Embeddings	Object2Vec, BlazingText Word2Vec

Essential Documentation¶

Core SageMaker Documentation¶

📖 SageMaker Developer Guide - Complete reference
📖 Built-in Algorithms - All algorithm details
📖 API Reference - API operations

Deep Dive Topics¶

📖 Distributed Training - Data and model parallelism
📖 SageMaker Pipelines MLOps - CI/CD for ML
📖 Feature Store - Online and offline feature storage
📖 Clarify Bias and Explainability - Detect bias, SHAP values
📖 Debugger - Debug training jobs

Hands-on Resources¶

SageMaker Examples GitHub - 100+ notebooks
AWS ML Blog - Latest techniques and case studies
SageMaker Studio Lab - FREE Jupyter environment

Study Strategy¶

Week 1-2: Data Engineering and EDA¶

AWS Glue, EMR, data formats (Parquet, RecordIO)
SageMaker Data Wrangler, feature engineering
Statistical analysis, handling imbalanced data
Hands-on: ETL pipeline, EDA in notebooks

Week 3-5: Modeling¶

SageMaker built-in algorithms (all of them!)
Deep learning frameworks (TensorFlow, PyTorch)
Hyperparameter tuning, distributed training
Evaluation metrics, cross-validation
Hands-on: Train XGBoost, Image Classification, DeepAR

Week 6-7: ML Implementation and Operations¶

Deployment options, multi-model endpoints
Model monitoring, SageMaker Pipelines
Security (VPC, IAM, encryption)
Cost optimization strategies
Hands-on: Deploy endpoint, set up monitoring, build pipeline

Week 8: Practice Exams and Review¶

Take 3+ full-length practice exams
Review incorrect answers thoroughly
Focus on weak areas
Memorize key formulas and decision trees

Recommended Resources¶

Official AWS Training¶

Exam Readiness: AWS Certified Machine Learning - Specialty - FREE on AWS Skill Builder

Practice Exams¶

AWS Official Practice Exam - $40 (highly recommended)
Tutorials Dojo - Comprehensive practice tests
Whizlabs - Multiple practice exams

Books and Courses¶

AWS Certified Machine Learning Specialty MLS-C01 by Frank Kane (Udemy)
AWS Machine Learning Specialty 2023 by Stephane Maarek (Udemy)

Hands-on Practice¶

SageMaker Free Tier - 250 hours Studio notebooks (first 2 months)
SageMaker Examples - Official GitHub repository with 100+ notebooks

Good luck with your AWS Machine Learning Specialty certification! 🚀