Test splitting a Random Forrest Classifier into a automated workflow

Take this notebook (https://github.com/dataquestio/project-walkthroughs/blob/master/football_matches/prediction.ipynb) and convert the model to something that can be deployed and iterated on.

Todo this I will separate the notebook into automated workflows:

• Data Pre-Processing
• Training
• Model Evaluation

Additional components will be a model registry. That way we can keep track of our models and dockerize models that exist in the registry.

mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./artifacts --host 0.0.0.0 then view at http://localhost:5000.

• Make diagram of flow
• Unit tests
• import parameterization of ML Flow Variables

Input:

{
    venue_code: Integer,
    opp_code: Integer,
    hour: Integer,
    day_code: Integer
}

Output: 0 or 1

1. preprocess.py Load data sources, format, transform. Then save new dataset

2. split.py Split the processed dataset into two, one to fit the model and one to test the model. We will train on previous seasons and predict on the current / latest season

3. train.py Load training data set and fit the model. save the model output

4. evaluate.py

Load a saved Model and test dataset for accuracy