wbx 0.1.0

W&B SDK extension to fetch runs for further analysis.

wbx

A Weight & Biases extension to fetch runs for further analysis.

Installation

pip install wbx

Usage

Get projects first

import wandb
from wbx import WBX


api = wandb.Api()
entity = "you-entity"

wbx = WBX(api, entity)
projects = wbx.list_projects(quick_see=True, max_val=10)

Then get runs from a specific project

project_name = "your-project-name"
wbx.list_configs(project_name)  # list all configs from runs in the project (for further filtering useless configs)

config_to_use = ["config1", "config2"]

runs_df = wbx.get_runs(
    project_name,
    group_keys=config_to_use,
    summary_keys=None,
    json_encode_non_scalars=False
)

After that, you can do analysis based on the runs_df DataFrame.

Or you can save it as a parquet file for later use.

wbx.to_parquet(runs_df, path="path/to/your/folder")

Contributing

Feel free to submit issues or pull requests.

License

MIT License. See LICENSE file for details.