After creating an experiment in PAI AutoML, you can monitor its progress in real time — including basic settings, trial results, hyperparameter combinations, and logs.
View basic information about an experiment
Log on to the PAI console.
In the left-side navigation pane, click Workspaces, then click the name of your workspace.
In the left-side navigation pane, choose Model Training > AutoML.
On the AutoML page, click the name of an experiment to open the Experiment Details page.
On the Basic Information tab, review the following sections:
Basic Settings — Shows the experiment ID and current status.
Trial Configuration, Execution Configurations, and Search Configurations — Show the settings you specified when creating the experiment.
Trial Status — Shows execution progress and a breakdown of trial statuses.
Experiment status reference
| Status | Description |
|---|---|
| CREATED | The system is creating a management process, parsing module configurations, and submitting tasks to Deep Learning Containers (DLC) or MaxCompute. Trials have not started yet. |
| RUNNING | All trials are running. |
| NO_MORE_TRIAL | The number of trials that have run in the experiment has reached the maximum you configured. No new trials are generated. After the last trial finishes, the status automatically changes to FINISHED. |
| TUNER_NO_MORE_TRIAL | The tuner has insufficient information to determine the next hyperparameter combination and has stopped generating new trials. |
| TERMINATING | You manually stopped the experiment. The experiment is in the process of stopping. |
| USER_CANCELED | You manually stopped the experiment and the experiment is stopped. |
| FINISHED | All trials have completed and at least one trial reached the FINISHED state. |
| FAILED | All trials have completed and all are in the FAILED state. This status also applies when you manually stop all trials. |
View trials
On the Experiment Details page, open the Trials tab to see all trials the experiment has generated.
Each trial represents one hyperparameter combination, generated automatically based on the algorithm you specified in Search Configurations. A trial runs as either a DLC job or one or more MaxCompute tasks, depending on the Execution Configurations of the experiment.
Trial status reference
| Status | Description |
|---|---|
| RUNNING | The trial is running. |
| FINISHED | The trial is completed. |
| FAILED | The trial failed. |
| USER_CANCELED | The trial was manually stopped. |
| EARLY_STOPPED | The trial was stopped because the EarlyStop configuration was triggered. |
Identify the best hyperparameter combination
Each trial has a Final Metric value, calculated as the weighted sum of metrics based on the metric weight configuration. Compare the Final Metric values across trials — the best combination is the one that optimizes in the direction specified by the Optimization parameter in Search Configurations.
View logs
Use logs to diagnose failures in an experiment or individual trial.
View experiment logs
On the Experiment Details page, click Logs in the upper-right corner. Alternatively, in the Basic Settings section of the Basic Information tab, click Logs next to Status.
In the Experiment Logs panel, review the log files:
Log file What it contains nnimanager.logExperiment management events: start, stop, and error messages dispatcher.logTask scheduling and resource management: trial allocation and resource allocation
View trial logs
On the Trials tab, find the trial you want to inspect and click Logs in the Actions column.
In the Trial Logs panel, review the log files:
Log file What it contains trial.logDetailed trial execution information stdoutStandard output: printed statements and debugging output from the trial stderrStandard error: exceptions, error messages, and warnings. Start here when a trial fails unexpectedly.