Lesson 4: SageMaker Built-in Algorithms

1. SageMaker Built-in Algorithms Overview

SageMaker provides 18+ built-in algorithms optimized to run distributed on AWS infrastructure. This is an extremely important topic in MLS-C01 — it typically accounts for 8-12 questions.

Exam tip: Memorize the "Problem Type → Algorithm" table. The exam always presents a scenario and asks for the appropriate algorithm. Key patterns: time series → DeepAR; anomaly → Random Cut Forest; NLP classification → BlazingText; tabular → XGBoost.

2. Supervised Learning Algorithms

Algorithm	Problem Type	Input	Key Trait
XGBoost	Classification, Regression	Tabular (CSV/LibSVM)	Top performer for tabular data, gradient boosting
Linear Learner	Binary/Multiclass classification, Regression	RecordIO, CSV	Fast, scalable, built-in regularization
Factorization Machines	Binary classification, Regression	RecordIO-protobuf (sparse)	Sparse data, recommendation systems, CTR prediction
KNN (k-Nearest Neighbors)	Classification, Regression	RecordIO-protobuf	Instance-based, no training, lazy learner
DeepAR	Time series forecasting	JSON Lines	Multiple related time series, probabilistic forecasts
Object2Vec	Embeddings	Paired sequences	Learn embeddings for words, products, users

3. NLP Algorithms

Algorithm	Output	Use Case
BlazingText	Word vectors or text classification	Sentiment analysis, spam detection, entity classification
Seq2Seq	Sequence → Sequence	Machine translation, summarization, Q&A
LDA (Latent Dirichlet Allocation)	Topics per document	Topic modeling, document categorization
NTM (Neural Topic Model)	Latent representations	Topic modeling with neural networks

Exam tip: BlazingText has 2 modes: (1) Word2Vec mode — unsupervised, generates word embeddings; (2) Text Classification mode — supervised, like FastText. Distinguish clearly when reading the question.

4. Unsupervised Learning Algorithms

Algorithm	Problem Type	Use Case
K-Means	Clustering	Customer segmentation, document grouping
PCA (Principal Component Analysis)	Dimensionality reduction	High-dimensional data, feature compression
Random Cut Forest (RCF)	Anomaly detection	Fraud detection, IoT anomaly, time series anomaly
IP Insights	Anomaly detection	Detect unusual IP-entity relationships, security

5. Computer Vision Algorithms

Algorithm	Task	Output
Image Classification	Multi-class classification	Class label + confidence
Object Detection	Locate + classify objects	Bounding boxes + labels
Semantic Segmentation	Pixel-level classification	Segmentation mask

6. Algorithm Selection Decision Tree

What is the problem type?
│
├── Tabular data, classification/regression?
│   └── XGBoost (best general choice)
│
├── Sparse features, recommendation, ad CTR?
│   └── Factorization Machines
│
├── Time series forecasting (multiple related series)?
│   └── DeepAR
│
├── Anomaly detection on time series / IoT?
│   └── Random Cut Forest (RCF)
│
├── Text classification / sentiment?
│   └── BlazingText (supervised mode)
│
├── Sequence-to-sequence (translation / summarization)?
│   └── Seq2Seq
│
├── Topic modeling?
│   └── LDA or NTM
│
├── Clustering?
│   └── K-Means
│
├── Dimensionality reduction?
│   └── PCA
│
└── Image tasks?
    ├── Classification only → Image Classification
    ├── Locate objects → Object Detection
    └── Pixel mask → Semantic Segmentation

7. Training Input Modes

Mode	How It Works	Best For
File Mode	Downloads entire dataset to training instance before starting	Small to medium datasets
Pipe Mode	Streams data directly from S3 during training	Very large datasets — no disk bottleneck
FastFile Mode	Access S3 as if local file system (via FUSE)	Random access patterns

Exam tip: When the question asks "reduce training time for large dataset", the answer is usually to switch to Pipe Mode with RecordIO format. Pipe Mode doesn't download the entire dataset — it streams directly from S3.

8. Cheat Sheet — Quick Reference

Keyword in Question	Algorithm
"tabular data", "structured data"	XGBoost
"time series", "forecast"	DeepAR
"anomaly detection"	Random Cut Forest
"recommendation", "sparse features"	Factorization Machines
"text classification", "sentiment"	BlazingText (supervised)
"word embeddings"	BlazingText (Word2Vec mode)
"translation", "summarization"	Seq2Seq
"topic modeling"	LDA or NTM
"clustering", "segmentation"	K-Means
"dimensionality reduction"	PCA
"bounding boxes", "object detection"	Object Detection
"pixel-level", "segmentation mask"	Semantic Segmentation
"IP address anomaly", "fraud login"	IP Insights

9. Practice Questions

Q1: A retail company wants to forecast product demand for the next 30 days across 5,000 product categories. Which SageMaker algorithm is BEST suited?

• A) K-Means
• B) Linear Learner
• C) DeepAR ✓
• D) Seq2Seq

Explanation: DeepAR is specifically designed for time series forecasting across multiple related time series. It learns global patterns from all 5,000 series simultaneously, providing probabilistic forecasts. This is exactly the use case it's optimized for.

Q2: An IoT system monitors server CPU usage. The team wants to detect unusual spikes automatically. Which SageMaker built-in algorithm should be used?

• A) XGBoost
• B) Random Cut Forest ✓
• C) BlazingText
• D) PCA

Explanation: Random Cut Forest (RCF) is SageMaker's built-in anomaly detection algorithm. It assigns an anomaly score to each data point and works well for time series anomaly detection, such as CPU usage spikes.

Q3: A data scientist is training a model on a 500 GB dataset. Training is very slow because downloading data to the training instance takes too long. Which change will MOST improve performance?

• A) Switch from CSV to JSON format
• B) Increase the training instance size
• C) Switch to Pipe Mode with RecordIO-protobuf format ✓
• D) Add more training epochs

Explanation: Pipe Mode streams data directly from S3 during training without downloading it first, eliminating the I/O bottleneck for large datasets. Combined with RecordIO-protobuf format, it dramatically reduces startup time.