---

From Training a Model to Storing It in S3 for Production Use

In the previous sections, we:

• set up SageMaker infrastructure

• configured IAM roles and Studio

• understood how SageMaker simplifies MLOps

Now we move to the next important step:

Creating a machine learning model and storing it in a location that SageMaker can use for deployment.

---

What We Are Building

We will:

• train a simple ML model (Iris classifier)

• save it as an artifact

• upload it to S3

• verify it

This is the foundation step before deployment.

---

Why This Step Matters

In real-world MLOps systems:

• models are not directly deployed from notebooks

• they are stored as versioned artifacts

• deployments always pull models from central storage (S3)

This ensures:

• reproducibility

• traceability

• version control

---

Architecture Overview

Notebook → Train Model → Save Artifact → Upload to S3 → Use for Deployment

---

Step 1 — Create an S3 Bucket

aws s3 mb s3://my-sagemaker-demo-bucket-abhishek

What this does

• Creates a storage location in AWS

• This bucket will store model artifacts

Think of S3 as:

The central storage for all ML models

---

Step 2 — Create IAM Execution Role

SageMaker needs permissions to:

• read/write S3

• run training jobs

• deploy models

---

Create Trust Policy

Save as trust.json

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": { "Service": "sagemaker.amazonaws.com" },
      "Action": "sts:AssumeRole"
    }
  ]
}

---

Create Role

aws iam create-role \
  --role-name SageMakerDemoExecutionRole \
  --assume-role-policy-document file://trust.json

---

Attach Permissions

aws iam attach-role-policy \
  --role-name SageMakerDemoExecutionRole \
  --policy-arn arn:aws:iam::aws:policy/AmazonSageMakerFullAccess

aws iam attach-role-policy \
  --role-name SageMakerDemoExecutionRole \
  --policy-arn arn:aws:iam::aws:policy/AmazonS3FullAccess

---

What this does

This role allows SageMaker to:

• access S3

• run jobs

• manage models

---

Step 3 — Find Default VPC and Subnets

---

Get Default VPC

aws ec2 describe-vpcs --filters Name=isDefault,Values=true \
  --query "Vpcs[0].VpcId" --output text

---

Get Subnets

aws ec2 describe-subnets --filters Name=vpc-id,Values=<VPC-ID> \
  --query "Subnets[*].SubnetId" --output text

Pick any two subnets.

---

What this does

• defines network where SageMaker runs

• ensures high availability

---

Step 4 — Create SageMaker Domain (Studio)

aws sagemaker create-domain \
  --domain-name demo-domain \
  --auth-mode IAM \
  --default-user-settings "ExecutionRole=arn:aws:iam::<ACCOUNT-ID>:role/SageMakerDemoExecutionRole" \
  --vpc-id <VPC-ID> \
  --subnet-ids "<SUBNET-1>" "<SUBNET-2>"

---

What this does

Creates:

• SageMaker Studio environment

• networking setup

• execution permissions

---

Save Domain ID

You will need it later.

---

Step 5 — Create User Profile

aws sagemaker create-user-profile \
  --domain-id <DOMAIN-ID> \
  --user-profile-name demo-user

---

What this does

• creates a workspace inside SageMaker

• each user gets isolated environment

---

Step 6 — Open SageMaker Studio

Go to:

AWS Console → SageMaker → Domains → demo-domain → Launch Studio

Select:

demo-user

---

What you will see

• JupyterLab interface

• ready-to-use ML environment

• no setup required

---

Step 7 — Train a Simple Model

Inside Studio, open a Python notebook.

---

Import Libraries

import boto3
import joblib
from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier
import os

---

Train Model

iris = load_iris()
X, y = iris.data, iris.target

model = DecisionTreeClassifier()
model.fit(X, y)

joblib.dump(model, "iris-model.pkl")

print("Model saved as iris-model.pkl")

---

What this does

• loads sample dataset

• trains a simple model

• saves model as file

---

Step 8 — Upload Model to S3

import boto3

s3 = boto3.client("s3")
bucket = "my-sagemaker-demo-bucket-abhishek"

s3.upload_file("iris-model.pkl", bucket, "model-artifacts/iris-model.pkl")

print("Uploaded to S3:", f"s3://{bucket}/model-artifacts/iris-model.pkl")

---

What this does

• uploads model artifact to S3

• creates a structured path

s3://bucket/model-artifacts/iris-model.pkl

---

Step 9 — Verify Upload

aws s3 ls s3://my-sagemaker-demo-bucket-abhishek/model-artifacts/

Expected output:

iris-model.pkl

---

What You Achieved

You successfully:

• trained a machine learning model

• saved it as an artifact

• uploaded it to S3

• verified its availability

---

Why This Is Important

This step is critical because:

• deployments always use stored artifacts

• models must be reproducible

• S3 acts as the source of truth

---

Real-World Flow

In production systems:

Training Pipeline → Model Artifact → S3 → Model Registry → Deployment → Endpoint

---

Key Takeaways

• Model artifacts must be stored centrally

• S3 is the default storage for SageMaker

• IAM roles control access

• Studio simplifies development

• This is the first step toward deploy

---