Resources

MLOps Roadmap
MLOps guide - Chip Huyen
MLOps with AWS - Bootcamp - Zero to Hero Series - Udemy
Mathematics-Basics to Advanced for Data Science And GenAI - Udemy

Architecture

Setting Up the AWS Account & Cloud9 for CLI

aws sts get-caller-identity
sudo yum -y install jq bash-completion

Variables for Hands On

export AWS_DEFAULT_REGION=$(curl -s 169.254.169.254/latest/dynamic/instance-identity/document | jq -r .region)
aws configure set default.region ${AWS_DEFAULT_REGION}
export AWS_ACCOUNT_ID=$(aws sts get-caller-identity --query "Account" --output text)
export VERIFY_ROLE_ARN="arn\:aws\:iam::${AWS_ACCOUNT_ID}\:role/MLOps"
export DATA_BUCKET="data-${AWS_DEFAULT_REGION}-${AWS_ACCOUNT_ID}"
export PIPELINE_BUCKET="mlops-${AWS_DEFAULT_REGION}-${AWS_ACCOUNT_ID}"
echo "export AWS_ACCOUNT_ID=$AWS_ACCOUNT_ID" >> ~/.bashrc
echo "export VERIFY_ROLE_ARN=$VERIFY_ROLE_ARN" >> ~/.bashrc
echo "export AWS_DEFAULT_REGION=$AWS_DEFAULT_REGION" >> ~/.bashrc
echo "export DATA_BUCKET=$DATA_BUCKET" >> ~/.bashrc
echo "export PIPELINE_BUCKET=$PIPELINE_BUCKET" >> ~/.bashrc

Default Setting

• Region : us-west-2

Setting Up the cli in cloud9

curl -O https://bootstrap.pypa.io/get-pip.py &&\
sudo python3 get-pip.py --user --no-warn-script-location &&\
rm get-pip.py &&\
python3 -m pip install -U pip boto3 numpy pandas wget awscli --user

wget https://github.com/fahimjason/mlops/raw/refs/heads/resources/mlops-tutorial.zip

unzip mlops-tutorial.zip

cd utils
ls
sudo chmod +x c9_resize.sh
./c9_resize.sh

1. Preperation:

Create S3 Bucket for Storage of dataset (Use Console if required)

aws s3 mb "s3://${DATA_BUCKET}" --region $AWS_DEFAULT_REGION

aws s3api put-bucket-versioning --bucket "${DATA_BUCKET}" --versioning-configuration Status=Enabled --region $AWS_DEFAULT_REGION

wget https://github.com/fahimjason/mlops/raw/refs/heads/resources/abalone.csv

Create S3 Bucket for Storage of output of Feature Engineering

aws s3 mb "s3://${PIPELINE_BUCKET}" --region $AWS_DEFAULT_REGION &&\
aws s3api put-bucket-versioning --bucket "${PIPELINE_BUCKET}" \
--versioning-configuration Status=Enabled --region $AWS_DEFAULT_REGION

Create GitHub Repository for Storage of code & artifacts

• Create Repository in GitHub with name mlops
• Clone to local system

git config --global user.name "<REPLACE_WITH_YOUR_GITHUB_USER_NAME>"
git config --global user.email "<REPLACE_WITH_YOUR_GITHUB_USER_EMAIL>"

# Set your GitHub username as a variable (Replace "your_github_username" with your GitHub username. Example: GITHUB_USERNAME="fahimjason")
GITHUB_USERNAME="your_github_username"

# Use the variable in the git clone command
cd ~/environment &&\
git clone https://github.com/$GITHUB_USERNAME/MLOps

Create Container Image Repository in Elastic Container Registry

• Create a Private Repository with repo name as abalone

Create Secret in AWS Secrets Manager

• Create a token for mlops Repository. If you don't know how to make it, click here
• Use the secret key as abalone and use the created token as value. Make the secret name as abalone-token

Validation

• Validate S3 Bucket for Data
• Validate S3 Bucket for Pipeline
• Validate Repository created under ECR

python3 ~/environment/utils/repository_validation.py

2. Configure the Assets

ETL assets

• Switch to a different branch

cd ~/environment/mlops
git checkout -b etl
cp ~/environment/etl/* .
ls

• Copy the files inside the etl folder to the mlops local repo
• Commit and push to codecommit

git add . &&\
git commit -m "initial commit of etl assets" &&\
git push --set-upstream origin etl

Training Assets

• Checkout to main branch

git checkout -b main

• Empty the branch

git branch --unset-upstream &&\
git rm -rf .

• Copy the model files from model folder

cp -R ~/environment/model/* .

• Modify the trainingjob.json as follows:
  • Replace Account id with respective Account id
  • Replace region with us-west-2
  • create a new service role for Sagemaker with name as MLOps

sed -i "s/<AccountId>/${AWS_ACCOUNT_ID}/" ~/environment/mlops/trainingjob.json
sed -i "s/<Region>/${AWS_DEFAULT_REGION}/" ~/environment/mlops/trainingjob.json

• push the code to main branch

git add . &&\
git commit -m "Initial commit of model assets" &&\
git push --set-upstream origin main

Service Quota Request

• ml.m5.xlarge, ml.m5.large for processing job usage - region = us-west-2

Create Docker Container - Train & Test

• Login to Public ECR Repo to enable creation of Containers

cd ~/environment/mlops &&\
aws ecr get-login-password --region $AWS_DEFAULT_REGION | docker login --username AWS --password-stdin 763104351884.dkr.ecr.$AWS_DEFAULT_REGION.amazonaws.com &&\
docker build --build-arg REGION=$AWS_DEFAULT_REGION -f Dockerfile -t tf_model:1.0 .

• build the Docker Image using Dockerfile

cd ~/environment/tests/unit_test/ && \
mkdir -p model && \
mkdir -p output && \
docker run --rm --name 'my_model' \
    -v "$PWD/model:/opt/ml/model" \
    -v "$PWD/output:/opt/ml/output" \
    -v "$PWD/input:/opt/ml/input" tf_model:1.0 train

Test of Prediction

• Prediction

docker run --rm --name 'my_model' \
-v "$PWD/model:/opt/ml/model" \
-v "$PWD/output:/opt/ml/output" \
-v "$PWD/input:/opt/ml/input" tf_model:1.0 test \
"[[4.400000000000000022e-01,3.449999999999999734e-01,1.000000000000000056e-01,3.659999999999999920e-01,1.219999999999999973e-01,9.049999999999999711e-02,1.199999999999999956e-01,0.000000000000000000e+00,1.000000000000000000e+00,0.000000000000000000e+00]]"

Serve on Port:8080

docker run --rm --name 'my_model' \
    -p 8080:8080 \
    -v "$PWD/model:/opt/ml/model" \
    -v "$PWD/output:/opt/ml/output" \
    -v "$PWD/input:/opt/ml/input" tf_model:1.0 serve

Test the API

cd ~/environment/tests/unit_test/ &&\
python app_test.py

System Test Assets

• Create new Branch

cd ~/environment/mlops && \
git checkout -b test

• Remove the files

git rm -rf .

• get the files for the branch

cp -R ~/environment/tests/system_test/* .

• Modify threshold value

sed -i "s/<Threshold>/3.1/" ~/environment/mlops/buildspec.yml

• push code to repo

git add -A &&\
git commit -m "Initial commit of system test assets" &&\
git push --set-upstream origin test

3. Pipeline Execution

Creation of pipeline

unset parameters
parameters="$parameters ParameterKey=ImageRepoName,ParameterValue=%s"
parameters="$parameters ParameterKey=ImageTagName,ParameterValue=%s"
parameters="$parameters ParameterKey=ModelName,ParameterValue=%s"
parameters="$parameters ParameterKey=RoleName,ParameterValue=%s"
parameters="$parameters ParameterKey=GitHubUser,ParameterValue=%s"

• Modify the mlops-pipeline.yml as follows:
  • Replace Secret name with abalone-token
  • Replace Secret Key with respective abalone

sed -i "s/<Token>/abalone-token/" ~/environment/pipeline/mlops-pipeline.yml
sed -i "s/<key>/abalone/" ~/environment/pipeline/mlops-pipeline.yml

cd ~/environment/pipeline && \
aws cloudformation package --template-file mlops-pipeline.yml \
--s3-bucket $PIPELINE_BUCKET --s3-prefix abalone-pipeline/artifacts \
--output-template-file mlops-pipeline-output.yml

aws s3 cp ~/environment/abalone.csv "s3://${DATA_BUCKET}/input/raw/abalone.csv" --region $AWS_DEFAULT_REGION

aws cloudformation create-stack --stack-name abalone-pipeline \
--template-body file://~/environment/pipeline/mlops-pipeline-output.yml \
--parameters $(printf "$parameters" "abalone" "latest" "abalone" "MLOps" $GITHUB_USERNAME) \
--capabilities CAPABILITY_NAMED_IAM CAPABILITY_AUTO_EXPAND

4. Review the Pipelines and Various Stages

• Load simulation

cd ~/environment/utils/ &&\
sed -i "s/<PipelineBucket>/${PIPELINE_BUCKET}/" load_sim.py

python3 load_sim.py

Thanks