wandb

Source code: tianshou/utils/logger/wandb.py

---

class WandbLogger(train_interval: int = 1000, test_interval: int = 1, update_interval: int = 1000, info_interval: int = 1, save_interval: int = 1000, write_flush: bool = True, project: str | None = None, name: str | None = None, entity: str | None = None, run_id: str | None = None, group: str | None = None, job_type: str | None = None, config: Namespace | dict | None = None, monitor_gym: bool = True, disable_stats: bool = False, log_dir: str | None = None)

Bases: BaseLogger

Weights and Biases logger that sends data to https://wandb.ai/.

This logger creates three panels with plots: train, test, and update. Make sure to select the correct access for each panel in weights and biases:

Example of usage:

logger = WandbLogger()
logger.load(SummaryWriter(log_path))
result = OnpolicyTrainer(policy, train_collector, test_collector,
                          logger=logger).run()

Parameters:

• train_interval – the log interval in log_train_data(). Default to 1000.

• test_interval – the log interval in log_test_data(). Default to 1.

• update_interval – the log interval in log_update_data(). Default to 1000.

• info_interval – the log interval in log_info_data(). Default to 1.

• save_interval – the save interval in save_data(). Default to 1 (save at the end of each epoch).

• write_flush – whether to flush tensorboard result after each add_scalar operation. Default to True.

• project (str) – W&B project name. Default to “tianshou”.

• name (str) – W&B run name. Default to None. If None, random name is assigned.

• entity (str) – W&B team/organization name. Default to None.

• run_id (str) – run id of W&B run to be resumed. Default to None.

• config (argparse.Namespace) – experiment configurations. Default to None.

• train_interval – the log interval in log_train_data(). Default to 1000.

• test_interval – the log interval in log_test_data(). Default to 1.

• update_interval – the log interval in log_update_data(). Default to 1000.

• info_interval – the log interval in log_info_data(). Default to 1.

• exclude_arrays – whether to exclude numpy arrays from the logger’s output

prepare_dict_for_logging(log_data: dict) → dict[str, int | Number | number | ndarray | float]

Prepare the dict for logging by filtering out invalid data types.

If necessary, reformulate the dict to be compatible with the writer.

Parameters:

log_data – the dict to be prepared for logging.

Returns:

the prepared dict.

load(writer: SummaryWriter) → None

write(step_type: str, step: int, data: dict[str, int | Number | number | ndarray | float]) → None

Specify how the writer is used to log data.

Parameters:

• step_type (str) – namespace which the data dict belongs to.

• step – stands for the ordinate of the data dict.

• data – the data to write with format {key: value}.

finalize() → None

Finalize the logger, e.g., close writers and connections.

save_data(epoch: int, env_step: int, gradient_step: int, save_checkpoint_fn: Callable[[int, int, int], str] | None = None) → None

Use writer to log metadata when calling save_checkpoint_fn in trainer.

Parameters:

• epoch – the epoch in trainer.

• env_step – the env_step in trainer.

• gradient_step – the gradient_step in trainer.

• save_checkpoint_fn (function) – a hook defined by user, see trainer documentation for detail.

restore_data() → tuple[int, int, int]

Restore internal data if present and return the metadata from existing log for continuation of training.

If it finds nothing or an error occurs during the recover process, it will return the default parameters.

Returns:

epoch, env_step, gradient_step.

static restore_logged_data(log_path: str) → dict[str, ndarray | dict[str, dict[str, ndarray | dict[str, TRestoredData]]]]

Load the logged data from disk for post-processing.

Returns:

a dict containing the logged data.