Wandb vs Tensorboard

[Programmer-RD-AI]

Wandb vs Tensorboard

Which is better ???

[mrdbourke]

TensorBoard comes baked into TensorFlow so it's great, however, Weights & Biases integrates very quickly into any code which uses TensorBoard-style logging and has a bunch more features.

I personally use both but go for Weights & Biases for larger experiments.

[Programmer-RD-AI]

OK Thank you so much @mrdbourke

[ShaidaMuhammad]

I prefer wandb, It's very flexible. You can use it with any sort of experimentation that involves logs.

[stolpa4]

I prefer wandb, It's very flexible. You can use it with any sort of experimentation that involves logs.

Why can't I use TensorboardX for any experimentation involving logs? I actually can, and it's quite convenient for that purpose.

However, you haven't provided any supporting evidence for your opinion.

The main point here is that Wandb is a much more feature-rich tool, offering dataset versioning, model management, and team collaboration features.

Moreover, Wandb is a cloud service. Although it may be possible to run it locally as a server, it requires additional effort. Therefore, if you only intend to use it without running it locally, you'll need internet access and a Wandb API key to run experiments on your server.

In contrast, TensorboardX is extremely simple. It saves everything to Tensorboard (tb) files locally, eliminating the need to set up any servers to save your logs, among other things. You can use it offline right out of the box. It is an open-source tool and promotes democratic usage.

Considering the points mentioned above, I personally prefer to use TensorboardX for my projects because I don't require the additional features provided by Wandb, and I don't appreciate its centralized, cloud-based approach.