torchtnt.utils.loggers.TensorBoardLogger

class torchtnt.utils.loggers.TensorBoardLogger(path: str, *args: Any, **kwargs: Any)

Simple logger for TensorBoard.

On construction, the logger creates a new events file that logs will be written to. If the environment variable RANK is defined, logger will only log if RANK = 0.

Note

If using this logger with distributed training:

• This logger should be constructed on all ranks
• This logger can call collective operations
• Logs will be written on rank 0 only
• Logger must be constructed synchronously after initializing the distributed process group.

Parameters:

• path (str) – path to write logs to
• *args – Extra positional arguments to pass to SummaryWriter
• **kwargs – Extra keyword arguments to pass to SummaryWriter

Examples:

from torchtnt.utils.loggers import TensorBoardLogger
logger = TensorBoardLogger(path="tmp/tb_logs")
logger.log("accuracy", 23.56, 10)
logger.close()

__init__(path: str, *args: Any, **kwargs: Any) → None

Methods

__init__(path, *args, **kwargs)
close()	Close writer, flushing pending logs to disk.
flush()	Writes pending logs to disk.
log(name, data, step)	Add scalar data to TensorBoard.
log_audio(*args, **kwargs)	Add audio data to TensorBoard.
log_dict(payload, step)	Add multiple scalar values to TensorBoard.
log_hparams(hparams, metrics)	Add hyperparameter data to TensorBoard.
log_image(*args, **kwargs)	Add image data to TensorBoard.
log_images(*args, **kwargs)	Add batched image data to summary.
log_scalars(main_tag, tag_scalar_dict[, ...])	Log multiple values to TensorBoard.
log_text(name, data, step)	Add text data to summary.

Attributes

path
writer