Running on Minikube

Install minikube according to the docs.

Start minikube

minikube start

Install k8s ingress

kubectl apply -f https://raw.githubusercontent.com/kubernetes/ingress-nginx/main/deploy/static/provider/cloud/deploy.yaml
minikube addons enable ingress

Build container image and publish to repository

chmod +x ./build.sh
./build.sh

To run the application using Helm

chmod +x ./deploy-helm.sh
./deplot-helm.sh

Add application url to /etc/hosts

echo "$(minikube ip) name-predict.local" | sudo tee -a /etc/hosts

Test the application

curl -X POST http://name-predict.local/predict \
    -H "Content-Type: application/json" \
    -d '{"name": "Quang", "top_n": 3}'

Expected response:

{
    "prediction":[
        {"value":-0.10589729249477386,"category":"Vietnamese"},
        {"value":-3.371195077896118,"category":"Chinese"},
        {"value":-3.7609355449676514,"category":"Korean"}
    ]
}

You should also be able to access the API documentation on http://name-prediction.local/docs

Explanation

• Created API using FastAPI
• Train script depended on the data to load the models
  • removed dependency by saving the parameters and loading only them
• FastApi should load the model in memory when it starts
  • Stops reading from disk with each request, keeps model in memory
  • Lower latency
• Could train model on K8s with jobs
  • Persistant volume to save artifacts
  • Prediction script could load from this volume
  • MLFlow for experiment tracking. Choose best or latest model

---

Name classification API

Task description

You are provided a set of scripts:

• scripts/train.py does model training
• scripts/predict.py does name classification based on the trained model

Your task is:

1. Expose the name classification via REST api endpoint. The endpoint should provide the ability to select top N most likely labels for the given name and should also provide the scores associated with each label.
2. Containerize the said API
3. Deploy the said container to k8s cluster using helm chart
4. Provide a document (readme) describing how to deploy and use the API.

You are free to use any REST API framework or library and design the endpoint as you see fit. You are free to duplicate and edit the code from scripts/ folder in a way you see would work best, as long as the classification can be run using your API. You are encouraged to use a local distribution of k8s like minikube.

Setup

pip install -r requirements.txt

Train the model

python scripts/train.py

This will save the weights file in the current directory. Training takes a few mins on CPU.

Run inference via CLI

Example of ruunning the script:

python scripts/predict.py Quang

Output will look like so

(-0.02) Vietnamese
(-4.66) Chinese
(-5.43) Korean