reference/generated/torchrl.record.loggers.RayLogger

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

RayLogger

torchrl.record.loggers.RayLogger(logger_cls: type[LoggerT], *args: Any, ray_actor_options: dict[str, Any] | None = None, ray_init_config: dict[str, Any] | None = None, **kwargs: Any) → None

Driver-owned Ray logger service with restricted worker clients.

Existing direct construction and use_ray_service=True continue to create this owner. Use client() before sending the logger to workers.

Parameters:

• logger_cls – Concrete Logger class.

• *args – Positional arguments forwarded to logger_cls.

• ray_actor_options – Options used to construct the Ray actor.

• ray_init_config – Options used to initialize Ray when needed.

• **kwargs – Keyword arguments forwarded to logger_cls.

Examples

>>> from torchrl.record import CSVLogger, RayLogger
>>> logger = RayLogger(CSVLogger, exp_name="run", log_dir="/tmp")  
>>> client = logger.client()  
>>> client.log_scalar("loss", 1.0, step=0)  
>>> logger.shutdown()