.. _header-n0:

Use Weights & Biases
====================

RecBole allows visualizing configs and metrics of different experiments
with W&B.

If you are new to W&B, please set up it first.

1. Start with a W&B account. `Create one now → <http://app.wandb.ai>`__

2. Go to your project folder in your terminal and install library:
   ``pip install wandb``

3. Inside your project folder, log in W&B: ``wandb login`` `your API
   key <https://wandb.ai/authorize>`__

You can start W&B in RecBole by passing ``--log_wandb=True`` as command
line argument, or use config dict. One can also turn ``log_wandb: True``
in the ``overall.yaml`` file or provide it as external config file.

**A Running Example:**

You can run BPR model on ml-100k dataset with W&B as follow:

.. code:: python

   python run_recbole.py --log_wandb=True

Then, go to your W&B project, you can see the following page, which
shows the change of metrics during the training and validation in each
epoch.

.. figure:: https://raw.githubusercontent.com/leoleojie/cloudimg/master/data/202202191625104.png
   :alt: 

You can also check the detailed configuration information and evaluation
metrics.

.. figure:: https://raw.githubusercontent.com/leoleojie/cloudimg/master/data/202202191625985.png
   :alt: 

W&B also allows you to compare these metrics and configs across
different experiments in the same project.

.. figure:: https://raw.githubusercontent.com/leoleojie/cloudimg/master/data/202202191648205.png
   :alt: 

.. figure:: https://raw.githubusercontent.com/leoleojie/cloudimg/master/data/202202191701276.png
   :alt: 

You can select different projects to experiment by modifying
``wandb_project`` parameter, which defaults to ``'recbole'``.

For more details about W&B, please refer to `Weights & Biases -
Documentation <https://docs.wandb.ai/>`__.