comet_ml.integration.kubeflow ¶

Inject the Comet Logger component which continuously track and report the current pipeline status to Comet.ml.

Parameters:

• api_key (str, default: None ) –

  Your Comet API Key, if not provided, the value set in the configuration system will be used.

• project_name (str, default: None ) –

  The project name where all of the pipeline tasks are logged. If not provided, the value set in the configuration system will be used.

• workspace (str, default: None ) –

  The workspace name where all of the pipeline tasks are logged. If not provided, the value set in the configuration system will be used.

1
2
3
4
5

@dsl.pipeline(name='ML training pipeline')
def ml_training_pipeline():
    import comet_ml.integration.kubeflow

    comet_ml.integration.kubeflow.comet_logger_component()

Logs the Kubeflow task identifiers needed to track your pipeline status in Comet.ml. You need to call this function from every components you want to track with Comet.ml.

Parameters:

• experiment (Experiment) –

  An already created Experiment object.

• workflow_uid (str) –

  The Kubeflow workflow uid, see below to get it automatically from Kubeflow.

• pod_name (str) –

  The Kubeflow pod name, see below to get it automatically from Kubeflow.

1
2
3
4
5
6
7
8

def my_component() -> None:
    import comet_ml.integration.kubeflow

    experiment = comet_ml.Experiment()
    workflow_uid = "{{workflow.uid}}"
    pod_name = "{{pod_name}}"

    comet_ml.integration.kubeflow.initialize_comet_logger(experiment, workflow_uid, pod_name)