Setup

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
pip install -r requirements-local.txt

Optionally

pip install -r requirements-mac.txt

You need to supply the following environment variables for the application to work. MLFLOW_TRACKING_USERNAME= MLFLOW_TRACKING_PASSWORD= MLFLOW_TRACKING_URI=

I suggest using a tool like dotenv for this.

Folder Structure

src

Contains the reusable application pieces:

• tooling to access the datasets
• the classifier implementations
• the experiment pipelines
• the service used in the feed.uvl project
• tests
• helpers

Experiments are started by running the shell scripts directly

Evaluation

Evaluation contains a set of jupyter notebooks that where used to create the evaluation.

Training

The models for the service are trained via the train.ipynb notebook in the project root.

The service container is build by running: docker build -t $CONTAINER_NAME -f "./Dockerfile" --build-arg mlflow_tracking_username=XXXXXX --build-arg mlflow_tracking_password=XXXXXX --build-arg mlflow_tracking_uri=XXXXXX . replacing XXXXXX with the appropriate credentials.