After you create an experiment, you can view the experiment details in real time, such as the basic information of experiments, trials, and execution and log details of each trial. This topic describes how to view the details of an experiment.
View basic information about an experiment
Go to the AutoML page.
Log on to the PAI console.
In the left-side navigation pane, click Workspaces. On the Workspaces page, click the name of the workspace that you want to manage.
In the left-side navigation pane, choose .
On the AutoML page, click the name of an experiment to go to the Experiment Details page.
On the Basic Information tab of the Experiment Details page, you can view the basic settings, trial configurations, execution configurations, search configurations, and trial status of the experiment.
In the Basic Settings section, you can view experiment information, such as ID and status. Valid values of experiment status:
CREATED: After you create an experiment, the system starts to create a management process, parse the configuration of each module, and submit tasks to Deep Learning Containers (DLC) or MaxCompute. The trials are not started when the experiment is in the CREATED state.
FINISHED: When all trials in an experiment are completed and at least one trial is in the FINISHED state, the experiment enters the FINISHED state.
FAILED: When all trials in an experiment are completed and are in the FAILED state, the experiment enters the FAILED state. If you manually stop all trials in an experiment, the experiment also enters the FAILED state.
RUNNING: When all trials in an experiment are in the RUNNING state, the experiment is in the RUNNING state.
TERMINATING: After you manually stop an experiment, the experiment enters the TERMINATING state.
USER_CANCELED: After you manually stop an experiment and the experiment is stopped, the experiment enters the USER_CANCELED state.
NO_MORE_TRIAL: When the number of trials that are run in an experiment reaches the maximum number of trials that you configured for the experiment, the experiment no longer generates additional trials. After the final trial is completed, the experiment status automatically changes from NO_MORE_TRIAL to FINISHED.
TUNER_NO_MORE_TRIAL nuner: When the experiment has insufficient information to determine the next combination of hyperparameters, the tuner stops generating additional trials. In this case, the experiment enters the TUNER_NO_MORE_TRIAL nuner state.
In the Trial Configuration, Execution Configurations, and Search Configurations sections, you can view the configurations of the experiment.
In the Trial Status section, you can view the execution progress and status statistics of the trial.
View trials
Go to the Experiment Details page. For more information, see View the basic information of an experiment.
On the Trials tab, you can view all trials that are automatically generated in the experiment.
An experiment generates various hyperparameter combinations based on the specified algorithm and creates a trial for each hyperparameter combination. A trial may be a DLC job or one or more MaxCompute tasks. The task type varies based on the execution configuration of the experiment. The system runs the trial based on the execution configuration.
View the execution details of a trial
On the Trials tab, you can view trial details, such as the execution status, final metrics, and the hyperparameter combinations.
Valid values of trial status:
RUNNING: The trial is running.
FINISHED: The trial is completed.
FAILED: The trial fails.
USER_CANCELED: The trial is manually stopped.
EARLY_STOPPED: The EarlyStop configuration of the trial is triggered and the trial is stopped.
Final Metric: The system uses the weighted sum value as the final metric based on the metric weight configuration.
Hyperparameter combination: The combination of hyperparameters that are automatically generated based on the specified algorithm. Each trial runs based on one hyperparameter combination.
Based on the Optimization parameter that you specified in the Search Configurations section, you can identify the optimal hyperparameter combination by comparing the Final Metric of each trial.
View logs
If an experiment or trial fails, you can identify the exceptions based on the logs.
View the logs of an experiment
Go to the Experiment Details page. For more information, see View the basic information of an experiment.
Click Logs on the upper-right part of the page. You can also click Logs to the right of Status in the Basic Settings section of the Basic Information tab.
In the Experiment Logs panel, view the details of the experiment log. The following types of logs are included:
nnimanager.log
The logs contain information about experiment management, such as the start, stop, and error messages of the experiment.
dispatcher.log
The logs contain information about task scheduling and resource management, such as trial allocation and resource allocation.
View the logs of a trial
Go to the Experiment Details page. For more information, see View the basic information of an experiment.
On the Trials tab, find the trial that you want to manage and click Logs in the Actions column.
In the Trial Logs panel, view the details of the trial logs.
The following types of logs are included:
trial.log
The logs contain detailed information about trial execution.
stdout
The logs contain the standard output of the trial, such as the printed statements and output, to help you understand the debugging information and execution results.
stderr
The logs contain the standard errors of the trial, such as exceptions, error messages, and warnings. If a trial does not run as expected, you can use the detailed information in stderr logs to troubleshoot and handle the issue.