mlflow_emissions_sdk

This package logs the carbon emissions of machine learning models.

Overview

Training and running industrial machine learning models requires significant computing resources. The environment impacts are often overlooked if not neglected. To make ML engineers more aware about the carbon footprint of their projects we created this package.

The package seamlessly integrates the CodeCarbon_package into the MLFlow MLOps package. Extending the logging capabilities of the latter with the carbon tracking methodology of the former.

Installation

To get started, use the package manager pip to install the mlflow-emissions-package:

pip install mlflow-emissions-package

Make sure that you have an mlflow client running:

mlflow ui

By default, the tracking uri is http://127.0.0.1:5000

To log your emissions you need to provide the tracker with the uri and the name of the experiment. If the experiment does not exist then the tracker will create a new experiment with the given name. Example

Usage

from pprint import pprint

import numpy as np
from sklearn.linear_model import LinearRegression

from mlflow_emissions_sdk.experiment_tracking_training import EmissionsTrackerMlflow




tracker_info = {
    "tracking_uri" : "http://127.0.0.1:5000",
    "experiment_name": "test_name"
}
# prepare training data
X = np.array([[1, 1], [1, 2], [2, 2], [2, 3]])
y = np.dot(X, np.array([1, 2])) + 3

# train a model
model = LinearRegression()

# Instatiates the tracker
runner = EmissionsTrackerMlflow()
runner.read_params(tracker_info)

# Starts the emissions tracking
runner.start_training_job()

history = model.fit(X, y)

# Ends the tracking
runner.end_training_job()

License

MIT