MLOps Ep.4 Productionizing Training Script (Train, Validate, and Save Models)

mlops-train-run-ml.py

# Get hold of the current run
run = Run.get_context()

# Experiment parameters
args = {
    "n_estimators": 120
}

reg_model = RandomForestRegressor(**args)
reg_model.fit(data["train"]["X"], data["train"]["y"])

# Validate Model on Validation Set
preds = reg_model.predict(data["test"]["X"])
mse = mean_squared_error(preds, y_test)
metrics = {"mse": mse}
print(metrics)

# Save Model
model_name = "sklearn_regression_model.pkl"
joblib.dump(value = reg_model, filename=model_name)

# Register a model
model = Model.register(model_path='sklearn_regression_model.pkl', model_name='sklearn_regression_model', workspace=ws)
print(model.name, model.id, model.version, sep='\t')
run.log('mean squared error', mse)