How to set experiment name such that it can be some unique name instead of version_0, ... etc.

[dazzle-me]

I'm currently running a lot of experiments and in order to track all of them in tensorboard I have to rename each experiment folder by hand (e.g. lightning_logs/version_0 -> lightning_logs/{unique_informative_exp_name}).
However, it would be better if I could pass the as an argument to Trainer.
I searched documentation thoroughly, but couldn't find anything related to names of experiments (I found only default_root_dir argument, however, it's not that interesting)

The question is follows :
Is there a way to set experiment name on program level? I'm using default logger.

[tshu-w]

You can pass Trainer a custom logger with the version specified.

from pytorch_lightning.loggers import TensorBoardLogger

logger = TensorBoardLogger("default_root_dir", version="your_version", name="my_model")
trainer = Trainer(logger=logger)

Here is the api of TensorBoardLogger

[dazzle-me]

Almost what I was looking for, but maybe it's even better to subdivide experiment directory into folders with model names too.
Thanks!

[tshu-w]

but maybe it's even better to subdivide experiment directory into folders with model names too.

You can change logger name to sub dir based on your model name 😄