MLOps

Design for an MLOps pipeline

Project Organization

├── LICENSE
├── Makefile           <- Makefile with commands like `make data` or `make train`
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── docs               <- A default Sphinx project; see sphinx-doc.org for details
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
│                         the creator's initials, and a short `-` delimited description, e.g.
│                         `1.0-jqp-initial-data-exploration`.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
│
├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
├── src                <- Source code for use in this project.
│   ├── __init__.py    <- Makes src a Python module
│   │
│   ├── data           <- Scripts to download or generate data
│   │   └── make_dataset.py
│   │
│   ├── features       <- Scripts to turn raw data into features for modeling
│   │   └── build_features.py
│   │
│   ├── models         <- Scripts to train models and then use trained models to make
│   │   │                 predictions
│   │   ├── predict_model.py
│   │   └── train_model.py
│   │
│   └── visualization  <- Scripts to create exploratory and results oriented visualizations
│       └── visualize.py
│
└── tox.ini            <- tox file with settings for running tox; see tox.readthedocs.io

---

Project based on the cookiecutter data science project template. #cookiecutterdatascience

Setting up the Project

Clone the repository

git clone https://github.com/StarsCDS/MLOps.git

Install dependencies

• Create a virtual environment using venv (you can also use conda instead of this)

python -m venv MLOps

• Install dependencies (you can also manually install all the dependencies from requirements.txt)

cd MLOps
make requirements

Data Versioning with dvc

• Pull the raw data (run pip install dvc dvc-gdrive if dvc command is not found)

dvc pull

Data versioning tips

• Add new data/modifications to data

dvc add <filepath>

• Push new data to remote

dvc push

• Version control the .dvc file using git

git add <filepath>.dvc

Data Preprocessing

• Unzip the raw data

make data

• Process the unzipped data

make features

• Train a model using the processed data

make train

• Predict an image using the trained model

make predict img='/path/to/image'

• Visualize the trained model

make visualization

Mlflow

• It is helpful for managing and monitoring machine learning experiments
• script.py has the mlflow code
• Experiments can be monitored by running the following commands (default: https://localhost:5000)

python script.py
mlflow ui

Github Actions

• It is helpful for automating all tasks related to code merging and deployment
• The configuration file for the action that is run at every push to main branch is at .github/workflows/python-app.yml
• It check for lint errors and runs unit tests before pushing code main branch and also on every pull request to the main branch

Containerization using Docker

• Build an image

docker build -t mlops-api .

• Bulid a container (run the image)

docker run -p 8000:8000 mlops-api.

Orchestration and Containerization with Kubernetes and Docker

minikube is being used to run kubernetes run on a single cluster for development purposes

Setting up

• Initialize minikube to make create an environment for kubernetes

minikube start

• List the docker images inside minikube

minikube image list

• Add the docker image inside the minikube VM

eval $(minikube docker-env)
docker build -t mlops-api .

• Create a deployment(container) in kubernetes

kubectl create deployment mlops-deploy --image=mlops-api

• Check if it's running properly

kubectl get deployment
kubectl get pod

• Expose the deployment

kubectl expose deployment mlops-deploy --type=NodePort --port=8000

• Check Node Port

kubectl get svc

• Run the service tunnel

minikube service mlops-deploy

• After this the api can be accessed

Tips

• Manually increase the replicas(number of containers)

kubectl scale deploy/mlops-deploy --replicas=5

• View logs of a particular container

kubectl logs -f <container-name>

References

• Deploy ML models on Kubernetes video
• Accessing Apps in Minikube
• FastApi in Containers

Further Reading

• Kubernetes
• Minikube
• FastApi

TODO

• Use cookiecutter/yeoman for project structure
• Add unit tests
• Add github actions for running the unit tests on pull requests
• Use dvc to manage data versions
• Use mlflow/kubeflow for mlops