Resources on how to use wandb docker

[jvisionvs]

Greetings:

After lots of research, I decided to go with wandb as the solution to several of the project management organization currently in place (or lack thereof).

With that, I am looking to acquire the most effective workflow using wandb docker and wandb local.

There is a page on this in the documents, and another in Github, but both are brief, and do not provide much information. All other documentation appears quite impressive (a major factor to me choosing this over the many other solutions). Whether featured by the company or a blog done by a third party, any good references to set up docker with wandb? Eventually, we will be spanning many parts of the data science pipeline (i.e., this is me doing a trial for a group at a company). So figured best practices and an efficient work environment should be set up first. Then, to start playing around with the outputs from the container runs to the project space (dashboard).

Any pointers, references, samples projects, or any other material that might be out there?

Thanks!

[a-sh0ts]

Hey @jvisionvs

I’m glad to hear about your decision to use wandb as the solution of choice for your organization.
I apologize for the lack of clarity in the documentation regarding wandb docker and wandb local . We are always updating our documentation so if there is something specific you would have liked to have seen about these commands please let us know!

wandb docker and wandb local are two different but heavily related container solutions to interacting with wandb .

” wandb docker is a command that starts a docker container, passes in wandb environment variables, mounts your code, and ensures wandb is installed. By default the command uses a docker image with TensorFlow, PyTorch, Keras, and Jupyter installed.“

wandb docker is intended to be used as pre-made docker container in which it makes it easy to interact with your wandb instance, with the necessary environment variables pre-configured and with the standard suite of deep-learning toolkits. You could as easily make your own docker container with a different base image by making sure to simply set the proper environment variables. This makes it so you have no need to directly login from your script to be able to log to your wandb instance. Here is one of the older articles we have outlining the usefulness of wandb docker.

Now I used the phrase “ wandb instance” because wandb local operates as your own wandb server. If the cloud solution does not suit your use case for a variety of reasons (be it the limitations set in place that can be adjusted, security reasons, etc.), you can spin up your own instance of wandb that would be exclusive to your organization. However, wandb local on localhost is good only for testing purposes. If you so decide to move with wandb local , you can use the documentation here to set up your instance for production workloads.

If you have any more questions please don’t hesitate to ask!
Happy Holidays,
Anish