23. AWS SageMaker for ML/AI

AWS SageMaker

---

Prerequisites

---

1. What is AWS SageMaker

AWS SageMaker is a fully managed machine learning (ML) platform that allows you to:

• build models
• train models
• deploy models

all in one place, without managing infrastructure

2. Why AWS SageMaker exists

In real-world ML, the problem is not just “making a model”.

The real challenges are:

• setting up servers (CPU / GPU)
• scaling training
• deploying models as APIs
• managing versions
• handling traffic

SageMaker solves these problems.

SageMaker = “ML + Infrastructure Automation”

Instead of this:

Write model → Setup server → Install dependencies → Train → Deploy API → Maintain server

You do this:

Write model → SageMaker handles everything else

3. What You Actually Use It For

3-1. Model Training

You can train models using:

• your own code (PyTorch, TensorFlow)
• built-in algorithms
• pre-trained models

without setting up GPU machines manually

3-2. Model Deployment

SageMaker can turn your model into:

• a REST API
• real-time prediction service

no need to build backend servers

3-3. Scalable Infrastructure

• automatically uses powerful instances
• can scale up/down depending on workload

useful for large datasets or production traffic

3-4. Experiment Management

• track training jobs
• manage different model versions

helps when experimenting with multiple models

flowchart LR
    A[Data] --> B[Training]
    B --> C[Model]
    C --> D[Endpoint]
    D --> E[Prediction API]

4. When You Should Use It

4-1. Good Use Cases

• production ML service
• large-scale training
• cloud-based deployment
• team collaboration

4-2. Not Necessary When

• simple experiments
• learning ML basics
• small datasets

5. How to create

5-1. Set Network

SageMaker is usually used on private subnect that can approach resource from air-gapped network for security. So we should build VPC, Subnet, NAT, IGW for the private network.

If it is not set, some aws servies like s3 can be approached

[SageMaker Studio]
        ↓
[Private Subnet]
        ↓
(NAT Gateway)
        ↓
Internet

How to build network in AWS

---

• How to build VPC : https://kcnote.github.io/posts/AWS-04-VPC/
• How to build Subnet : https://kcnote.github.io/posts/AWS-05-Subnet/
• How to build IGW : https://kcnote.github.io/posts/AWS-06-Internet-Gateway/
• How to build Route : https://kcnote.github.io/posts/AWS-07-Route-Table/
• How to build NAT : https://kcnote.github.io/posts/AWS-08-NAT-Gateway/

---

5-2. Search SageMaker

5-3. Click Navigation pane → “Domains”

5-4. Click Button → “Create Domain”

5-5. Create an Amazon SageMaker Unified Studio domain

5-6. Enter Domain management

5-7. Create Project on Domain

5-8. Connect Project

5-9. Create Editor VS Code

If you want to create VS Code Editor, you should use instance over 8GB RAM without t instance type.