Skip to content

Advanced Usage

Using Custom Loggers

By default, Tensorboard logging is enabled, however, you can easily switch to WandB (W&B) (after registering and logging in on your system) via a command line override.

Login to W&B

To set-up W&B login credentials on your system, run the following and enter the API Key (find your API Key here):

wandb login

wandb: Logging into wandb.ai. (Learn how to deploy a W&B server locally: https://wandb.me/wandb-server)
wandb: You can find your API key in your browser here: https://wandb.ai/authorize?ref=models
wandb: Paste an API key from your profile and hit enter, or press ctrl+c to quit:
wandb: No netrc file found, creating one.
wandb: Appending key for api.wandb.ai to your netrc file: /home/users/{USERNAME}/.netrc
wandb: Currently logged in as: {USER_NAME} ({WANDB_USER}) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin

This will store your API key locally under:

~/.netrc

machine api.wandb.ai
  login user
  password b???...

Use W&B Logger

Now, the W&B logger can be used for training.

canari_ml train -cd configs/train/ -cn custom_train.yaml +logger=wandb

Alternatively, you could also define this in your training YAML config file as a config group after loading the default /train config file.

# @package _global_

defaults:
  - ../preprocess/train_1979_ua700_3days.yaml
  - /train
  - logger: wandb   # (1)!
  - _self_
  1. Must be placed after loading the default /train, but any order after it.

Specify Accelerators

By default, the auto keyword is used to allow PyTorch Lightning to auto-select the accelerator, and a single accelerator device is set to be used. To override which accelerator to use, and how many GPUs (or which GPU by GPU id):

canari_ml train -cd configs/train/ -cn custom_train.yaml trainer.accelerator="gpu" strategy="ddp" trainer.devices="[0]"

This will run on a single GPUs, with id of 0.

Refer to the Lightning docs for more details.