Initialize a sweep

After you define a sweep configuration and before you launch agents, initialize a sweep to register its configuration with Weights & Biases and receive a sweep ID. Agents use the sweep ID to fetch run instructions. After a sweep is initialized, it’s ready to coordinate hyperparameter searches across one or more machines. Weights & Biases uses a sweep controller to manage sweeps across one or more machines. After a run completes, the controller issues a new set of instructions that describe the next run to execute, and agents pick up these instructions and run them. By default, Weights & Biases hosts the controller, and agents run on your machines. If your configuration includes scheduler, the optimization engine runs in your own environment as a separate process. Start that process after you initialize the sweep. See Configure an optimization engine for engine selection and dependencies. Before you initialize a sweep, you must have a sweep configuration defined either in a YAML file or a nested Python dictionary object in your script. For more information, see Define sweep configuration. The following code snippets show how to initialize sweeps with the CLI and within a Jupyter notebook or Python script. Choose the method that matches your workflow.

The sweep and the run must be in the same project. The project name you provide when you initialize Weights & Biases (wandb.init()) must match the project name you provide when you initialize a sweep (wandb.sweep()).

Python script or notebook
CLI

Use the W&B SDK to initialize a sweep. Pass the sweep configuration dictionary to the sweep parameter. Optionally set the project parameter to the project name where you want Weights & Biases to store the output of the run. If you don’t specify the project, Weights & Biases puts the run in an “Uncategorized” project.

import wandb

# Example sweep configuration
sweep_configuration = {
    "method": "random",
    "name": "sweep",
    "metric": {"goal": "maximize", "name": "val_acc"},
    "parameters": {
        "batch_size": {"values": [16, 32, 64]},
        "epochs": {"values": [5, 10, 15]},
        "lr": {"max": 0.1, "min": 0.0001},
    },
}

sweep_id = wandb.sweep(sweep=sweep_configuration, project="project-name")

The wandb.sweep() function returns the sweep ID. The sweep ID includes the entity name and the project name. Note the sweep ID, which you pass to agents when you start them. With the sweep ID, you’re ready to launch one or more agents to execute runs.

Use the Weights & Biases CLI to initialize a sweep. Provide the name of your configuration file. Optionally set --entity and --project to choose the entity and project. If you don’t specify the project, Weights & Biases puts the run in an “Uncategorized” project.Use the wandb sweep command to initialize a sweep. The following code example initializes a sweep for a sweeps_demo project and uses a config.yaml file for the configuration.

wandb sweep --project sweeps_demo config.yaml

This command prints out a sweep ID. The sweep ID includes the entity name and the project name. Note the sweep ID, which you pass to agents when you start them. With the sweep ID, you’re ready to launch one or more agents to execute runs.If the sweep configuration includes scheduler key, the command also prints the wandb sweep-scheduler command to use for this sweep.

Next, start the scheduler and sweep agents. Start a scheduler only if your configuration includes scheduler.

Weights & Biases

Guides

Integrations

Reference

Support