sweeps

0.2.0 verified Fri May 01 auth: no python

Weights and Biases Hyperparameter Sweeps Engine. Current version 0.2.0. Release cadence is irregular; last release was v0.2.0.

pip install sweeps

Common errors

error ModuleNotFoundError: No module named 'sweeps' ↓

cause The `sweeps` library is not installed.

fix

Run pip install sweeps to install the package.

error wandb: ERROR Error while calling wandb sweep function: ... ↓

cause The training function defined for the sweep agent may have an error, or required packages are missing.

fix

Check the training function for bugs and ensure all dependencies (e.g., scikit-learn) are installed.

Warnings

gotcha The `sweeps` library is a standalone engine but sweep execution still requires `wandb` to start the agent. Do not attempt to run sweeps without `wandb` installed. ↓

fix Ensure `wandb` is installed: `pip install wandb`

deprecated Sweep methods `bayes` and `grid` may not be fully supported in future versions. Prefer `random` or custom search. ↓

fix Use `method: 'random'` in sweep configuration.

Imports

sweeps
```
import sweeps
```

Quickstart

Initialize a W&B sweep with sweeps config, define training function, launch agent.

import wandb
import sweeps

sweep_config = {
    'method': 'random',
    'parameters': {
        'learning_rate': {'min': 0.001, 'max': 0.1},
        'batch_size': {'values': [16, 32, 64]}
    }
}

sweep_id = wandb.sweep(sweep_config, project='my-sweep')
def train():
    run = wandb.init()
    lr = wandb.config.learning_rate
    bs = wandb.config.batch_size
    # ... training code ...
    wandb.log({'accuracy': 0.9})

wandb.agent(sweep_id, function=train, count=5)