Runs

A single unit of computation logged by W&B is called a run. You can think of a W&B run as an atomic element of your whole project. You should initiate a new run when you:

• Train a model
• Change a hyperparameter
• Use a different model
• Log data or a model as a W&B Artifact
• Download a W&B Artifact

For example, during a sweep, W&B explores a hyperparameter search space that you specify. Each new hyperparameter combination created by the sweep is implemented and recorded as a unique run.

tip

Some key things to consider when you create and manage runs:

• Anything you log with wandb.log is recorded in that run. For more information on how log objects in W&B, see Log Media and Objects.
• Each run is associated to a specific W&B project.
• View runs and their properties within the run's project workspace on the W&B App UI.
• There is only at most one active wandb.Run in any process, and it is accessible as wandb.run.

Create a run

Create a W&B run with wandb.init():

import wandb

run = wandb.init()

We recommend you specify a project name and a W&B entity when you create a new run. W&B creates a new project (if the project does not already exist) within the W&B entity you provide. If the project already exists, W&B stores the run in that project.

For example, the following code snippet initializes a run that is stored in a project called model_registry_example that is scoped within a wandbee entity:

import wandb

run = wandb.init(entity="wandbee", \
        project="model_registry_example")

W&B prints the name of the run that is created along with a URL path to find out more information about that specific run.

For example, the code snippet above produces this output:

Organize runs with run names and run IDs

By default, W&B generates a random name and run ID when you initialize a new run.

In the preceding example, W&B generates a run name called likely-lion-9 and a run ID of xlm66ixq. The likely-lion-9 run is stored in a project called model_registry_example.

note

Run names that are generated by W&B are not guaranteed to be unique.

You can provide a unique run ID identifier with the id parameter and a name for your run with the name parameter when you initialize a run. For example,

import wandb

run = wandb.init(
    entity="<project>", 
    project="<project>", 
    name="<run-name>", 
    id="<run-id>"
)

Use run names and run IDs to quickly find your experiments in your project within the W&B App UI. You can find detailed information about a specific run with the URL:

W&B App URL for a specific run

https://wandb.ai/entity/project-name/run-id

Where:

• entity: The W&B entity that initialized the run.
• project: The project where the run is stored.
• run-id: The run ID for the run.

tip

W&B suggests that you specify a project name when you initialize a run. If a project is not specified, W&B stores runs in a project called "Uncategorized".

See the wandb.init reference documentation for a full list of parameters you can use.

View a run

View a specific run within the project the run was logged to:

1. Navigate to the W&B App UI at https://wandb.ai/home.
2. Navigate to the W&B project you specified when you initialized the run.
3. Within your project's workspace, you will see a table labeled Runs. This table lists all the runs that are in your project. From the list of runs shown, select the run you want to view.
4. Next, select the Overview Tab icon.

The following image demonstrates information about a Run called sparkling-glade-2:

The Overview Tab shows the following information about the run you selected:

• Run name: The name of the run.
• Description: A description of the run that you provided. This field is left initially blank if no description was specified when you create the run. You can optionally provide a description for the run with the W&B App UI or programmatically.
• Privacy: Privacy settings of the run. You can set it to either Private or Public.
  • Private: (Default) Only you can view and contribute.
  • Public: Anyone can view.
• Tags: (list, optional) A list of strings. Tags are useful for organizing runs together, or applying temporary labels like "baseline" or "production".
• Author: The W&B username that created the run.
• Run state: The state of the run:
  • finished: script ended and fully synced data, or called wandb.finish()
  • failed: script ended with a non-zero exit status
  • crashed: script stopped sending heartbeats in the internal process, which can happen if the machine crashes
  • running: script is still running and has recently sent a heartbeat
• Start time: The timestamp when the run started.
• Duration: How long, in seconds, the run took to finish, fail, or crash.
• Run path: The unique run identifier. It has the form of entity/project/run-ID
• Host name: Where the run was launched. The name of your machine is displayed if you launched the run locally on your machine.
• Operating system: The operating system used for the run.
• Python version: The Python version used for the run.
• Python executable: The command that started the run.
• System Hardware: The hardware used to create the run.
• W&B CLI version: The W&B ClI version installed on the machine that hosted the run command.
• Job Type:

Below the overview section, you will additionally find information about:

• Artifact Outputs: Artifact outputs produced by the run.
• Config: List of config parameters saved with wandb.config.
• Summary: List of summary parameters saved with wandb.log(). By default, this value is set to the last value logged.

For more information about how to organize multiple Runs in a project, see the Runs Table documentation.

For a live example of a Project's Workspace, see this example project.

End a run

W&B automatically ends runs and logs data from that run to your W&B project. You can end a run manually with the run.finish command. For example:

import wandb

run = wandb.init()
run.finish()

info

W&B suggests that you use the wandb.finish method at the end of the child process if you call wandb.init from a child process.