artefactory/xhec-mlops-project-solution

This repository contains the solution for the X-HEC MLOps Project on the industrialization of Abalone age prediction model.

xhec-mlops-project-solution

This repository contains the solution for the X-HEC MLOps Project on the industrialization of Abalone age prediction model.

Table of Contents

• xhec-mlops-project-solution
  • Table of Contents
  • Installation
  • Usage
  • Training the model

Installation

Build the image docker with the following command:

docker build -t abalone:solution -f Dockerfile.app .

Usage

1. Run the prediction API:

docker run -d -p 0.0.0.0:8000:8001 -p 0.0.0.0:4200:4201 abalone:solution

Note

The -d flag is used to run the container in detached mode. The container will thus run in the background.

If you want to get the logs of the container, you can:

i. Get the container ID:

docker ps

ii. Copy/paste the container ID

iii. Get the logs:

docker logs <container_id> --follow

2. Go to http://localhost:8000/docs

3. In the /predict section, click on "Try it out".

4. Replace the Request body with the data of your choice.

Example:

[
    {
        "sex": "M",
        "length": 0.455,
        "diameter": 0.365,
        "height": 0.095,
        "whole_weight": 0.514,
        "shucked_weight": 0.2245,
        "viscera_weight": 0.101,
        "shell_weight": 0.15
    },
    {
        "sex": "M",
        "length": 0.35,
        "diameter": 0.265,
        "height": 0.09,
        "whole_weight": 0.2255,
        "shucked_weight": 0.0995,
        "viscera_weight": 0.0485,
        "shell_weight": 0.07
    }
]

5. Click on "Execute" to get the prediction. You should get a 201 response with the prediction in the Response body, like this:

{
  "predicted_abalone_ages": [
    9.28125,
    7.90625
  ]
}

6. You can also visualize the Prefect Flow Runs here: http://localhost:4200/flow-runs

7. When done with the API, you can stop the container with:

docker kill <container_id>

Training the model

This repository comes with a pre-trained model and pre-fitted encoder (src/web_service/local_objects). If you want to train the model yourself, you can:

1. Install the dependencies in a virtual environment (python 3.9 or higher):

pip install -r requirements.txt
pip install -r requirements-dev.txt

1. Put your data in the data folder. The data should be a CSV file with the same columns as the Abalone dataset.

2. Run the training script:

python3 src/modelling/main.py data/abalone.csv

This command will override the pre-trained model and encoder in the src/web_service/local_objects folder.

You can finally rebuild the docker image and run it again to use your new model.