dplutils.observer.mlflow.MlflowObserver

dplutils.observer.mlflow.MlflowObserver#

class dplutils.observer.mlflow.MlflowObserver(run=None, experiment=None, tracking_uri=None, **mlflow_kwargs)[source]#

Mlflow-based observer

MLflow is an ML experiment tracker with included explorer UI. See mlflow/mlflow for details.

Parameters:

  • run – Existing mlflow run object to use for tracking. In this case it is assumed the tracking_uri is that set globally.

  • experiment – Name of experiment under which to create run (if run not supplied).

  • tracking_uri – tracking uri, e.g. file://... or mlflow://, etc. See mlflow docs for details.

  • mlflow_kwargs – In case an existing

  • supplied (run is not) – created, in which case mlflow_kwargs (excluding experiment_id) will be passed to its instantiation, using mlflow.MlflowClient.create_run.

  • be (one will) – created, in which case mlflow_kwargs (excluding experiment_id) will be passed to its instantiation, using mlflow.MlflowClient.create_run.

__init__(run=None, experiment=None, tracking_uri=None, **mlflow_kwargs)[source]#

Methods

__init__([run, experiment, tracking_uri])

increment(name[, value])

Increment a counter

observe(name, value, **kwargs)

Observe a metric value

param(name, value, **kwargs)

Record a run parameter

timer(name, **kwargs)

Return a timer context manager recoding to this observer
Contents