The objective of this project is to be able to predict how highly rated a coffee would be on Coffee Review, based upon some meta-data about the selected coffee.

The dataset can be download from Kaggle. You can either download this:

• Directly from the website
• Using the Kaggle API as follows:

  !kaggle datasets download -d schmoyote/coffee-reviews-dataset -f simplified_coffee.csv -p data

To train a linear regression model, run the following script:

training/train.py

The model is served within a separate Docker container. This is hosted using AWS Lambda.

To build the server locally, run the following command:

docker build server -t coffee-prediction-server:latest

To start the server for the first time, use the following command:

docker run -it --network coffee-rating-prediction_devcontainer_default --hostname coffee_server --name coffee-prediction-server coffee-prediction-server:latest

To restart the server, run the following:

docker start -i coffee-prediction-server

Then run the following to test the server:

python test_predict.py

To investigate the predicted rating of a given coffee, you can use the interactive streamlit dashboard. This is hosted on Streamlit Cloud.

You can also launch the dashboard locally by running the following command:

streamlit run dashboard.py