Overview

When calling tracking module in-cluster, Polyaxon performs a couple of checks to automatically load a context defined by the scheduler. Additionally, this context contains a scoped token to communicate with the API to only authorize the job or service to perform requests to accessible entities only.

In-cluster tracking

If you are running an experiment in-cluster:


from random import random, randint

from polyaxon.schemas import V1ArtifactKind
from polyaxon import tracking

if __name__ == "__main__":
    # No need to specify anything, a context is provided in-cluster
    tracking.init()

    # Log additional parameter
    tracking.log_inputs(param1=randint(0, 100))

    # Log additional tags
    tracking.log_tags(["foo", "bar"])

    # Log a single metric
    tracking.log_metric(name="metric1", value=random(), step=OPTIONAL, timestamp=OPTIONAL)
    # Log multiple metrics at once
    tracking.log_metrics(metric2=random(), metric3=random(), step=OPTIONAL, timestamp=OPTIONAL)

    # Log a final results
    tracking.log_outputs(res1=randint(0, 100), res2="test value", ...)

    # Save the artifact
    asset_path = tracking.get_outputs_path("test.txt")
    with open(asset_path, "w") as f:
        f.write("Artifact content.")
    # Track the lineage Name will default to test
    tracking.log_artifact_ref(path=asset_path, kind=V1ArtifactKind.FILE)

    # Save a second artifact
    asset_path = tracking.get_outputs_path("file.csv")
    with open(asset_path, "w") as f:
        f.write("Artifact content.")
    # Track the with a different name
    tracking.log_artifact_ref(path=asset_path, kind=V1ArtifactKind.CSV, name="my-csv")