After creating an experiment, you can manage both the experiment and its trials throughout the experiment lifecycle. Supported operations include modifying, cloning, stopping, and deleting experiments, as well as rerunning failed trials.
Experiment and trial states
| State | Applies to | Description |
|---|---|---|
| RUNNING | Experiment, trial | The experiment or trial is actively executing. |
| FAILED | Trial | The trial completed with errors. |
Prerequisites
Before you begin, ensure that you have:
A PAI workspace with AutoML enabled
An experiment created in the workspace
Modify an experiment
While an experiment is in the RUNNING state, modify its parameters to steer the search toward better hyperparameter combinations or find the optimal combination faster. Results vary depending on which parameters you change.
Log on to the PAI console.
In the left-side navigation pane, click Workspaces. On the Workspaces page, click the name of the target workspace.
In the left-side navigation pane, choose Model Training > AutoML.
Open the experiment using either of the following methods:
If the experiment is in the RUNNING state, click Edit in the Actions column of the target experiment.
Click the experiment name to open the Experiment Details page, then click Edit in the upper-right corner.
On the Update Experiment page, modify the parameters and click Submit.
Rerun a trial
If an experiment is in the RUNNING state and a trial within it is in the FAILED state, rerun the trial to retry the failed trial's configuration.
Rerunning a trial creates a new trial. This starts a Deep Learning Containers (DLC) job or MaxCompute task and consumes cloud computing resources.
On the AutoML page, click the experiment name to open the Experiment Details page.
On the Trials tab, find the FAILED trial and click Rerun in the Actions column.
In the confirmation message, click OK.
Stop a trial
Stop a RUNNING trial to free up resources without affecting the overall status of the experiment.
On the AutoML page, click the experiment name to open the Experiment Details page.
On the Trials tab, find the RUNNING trial and click Stop in the Actions column.
Stop an experiment
On the AutoML page, click Stop in the Actions column of a RUNNING experiment to stop it and free up resources. The system stops all RUNNING trials before stopping the experiment itself.
Clone an experiment
Clone an experiment to create an identical experiment, or modify the experiment parameters based on the existing experiment to create a new experiment. On the AutoML page, open the clone dialog using either of the following methods:
Click Clone in the Actions column of the target experiment.
Click the experiment name to open the Experiment Details page, then click Clone in the upper-right corner.
On the Create Experiment page, review or adjust the parameters and click Submit.
Delete an experiment
You can only delete experiments that are not in the RUNNING state. Stop the experiment first if it is currently running.
After you delete an experiment, you cannot view the experiment information or find the experiment data. Proceed with caution.
On the AutoML page, delete the experiment using either of the following methods:
Click Delete in the Actions column of the target experiment. In the Note message, click OK.
Click the experiment name to open the Experiment Details page, then click Delete in the upper-right corner. In the Note message, click OK.
Search for experiments
Filter experiments on the AutoML page using the following criteria:
| Filter | How to use |
|---|---|
| Experiment ID or name | Enter a partial or full experiment ID or name in the search box. |
| Creator | Enter the Alibaba Cloud account ID of the experiment creator in the Creator field. |
| Status | Select a status to show only experiments in that state. |