The experiments table is used to visualize all the experiments logged to a given project, as well as the metadata, parameters, and metrics logged to each of those experiments. Users can filter and sort the data in the table to easily explore and interpret results logged to the experiments.
When present in a
rubicon_ml.viz.Dashboard, an experiments table acts a controller to select the experiments visible in the other widgets in the dashboard.
The table itself is a Dash
data_table. More information can be found in the Dash documentation.
import random from rubicon_ml import Rubicon from rubicon_ml.viz import ExperimentsTable
First, we’ll create a few experiments and log some parameters and a metric to them.
rubicon = Rubicon(persistence="memory", auto_git_enabled=True) project = rubicon.get_or_create_project("experiment table") for i in range(0, 24): experiment = project.log_experiment() experiment.log_parameter(name="max_depth", value=random.randrange(5, 25, 5)) experiment.log_parameter(name="n_estimators", value=random.randrange(2, 12, 2)) experiment.log_metric(name="accuracy", value=random.random())
Now, we can instantiate the
ExperimentsTable object with the experiments we just logged and view the table right in this notebook with
show. The Dash application itself will be running on http://127.0.0.1:8050/ when running locally. Use the
serve command to launch the server directly without rendering the widget in the current Python interpreter.
ExperimentsTable( experiments=project.experiments(), ).show()
Dash is running on http://127.0.0.1:8050/