Creating and Managing Experiments (Beta)

Overview

Experiment management provides algorithm engineers with a unified workspace to systematically organize, track, and compare multiple training tasks. You can assign a training job to an experiment for easy classification and management. Here are two ways to use experiments:

1. Specify an experiment when creating a training job (you can either create a new experiment alongside the job or select an existing one).
2. When cloning an existing training job, it will be included by default in the experiment where the source job resides.

Constraints

1. The experiment feature in the new console is currently in beta, whereas creating experiment projects in the old console is a commercial feature. For details about how to switch between the old and new consoles, see New and Old Consoles.
2. This section describes how to create and manage experiment projects based primarily on the new console.

Creating an Experiment Project

1. Log in to the ModelArts console.
2. In the navigation pane, choose Model Build > Experiments.
   Figure 1 Experiments
   

3. Click Create Experiment in the upper right corner. In the dialog box that is displayed, enter the experiment project information and click OK. Table 1 describes the parameters.

   Table 1 Parameters for creating an experiment

   Parameter	Description
   Experiment Name	Experiment name. This parameter is mandatory. Naming rules: The value can contain 1 to 128 characters. Only letters, digits, hyphens (-), and underscores (_) are allowed.
   Description (Optional)	Experiment description, which helps you learn about the experiment information in the experiment list. Enter 0 to 512 characters. Angle brackets (<>) are not allowed.

Viewing and Managing Experiments

All experiment projects are listed in the experiment workspace. You can view experiment details including Name/ID, Training Jobs, Created, Modified, Description, and Operation. Table 2 describes the related information.

Note: Experiments can be searched in the search bar by name, ID, total training jobs, creation time, modification time, or description.

Viewing Experiment Details

In the experiment workspace, click an experiment name to view its detailed information. Table 3 lists the viewable information.