Set Up Experiment Tracking Stack

Experiment tracking can help you better understand your model performance during machine learning model training. envd already has built-in features to help you integrate with your favorite tools.

We will use TensorBoard as an example.

def tensorboard(port):
    runtime.mount(host_path="/var/log/tensorboard", envd_path="/home/envd/logs")
    runtime.daemon(commands=[
        ["tensorboard", "--logdir", "/home/envd/logs", "--port", "8888", "--host", "0.0.0.0", ">>tensorboard.log", "2>&1"],
    ])
    runtime.expose(envd_port=port, host_port=port, service="tensorboard")


def build():
    base(os="ubuntu20.04", language="python")
    install.python_packages(name=["tensorboard"])
    tensorboard(8888)

Make sure you have permission for the mounted directory.

This example includes mount, daemon and expose. For more details, can check envd API reference.