TensorBoard SummaryWriter

Utilities for logging data to TensorBoard.

---

SummaryWriter

skrl provides a simple SummaryWriter implementation to log data to TensorBoard without relying on third-party libraries.

API

class skrl.utils.tensorboard.SummaryWriter(log_dir: str, *, queue_size: int = 10, flush_seconds: int = 120)

Bases: object

TensorBoard summary writer.

Parameters:

• log_dir – Directory to save the event file.

• queue_size – Maximum number of events to keep in the queue before forcing a flush to persistent storage.

• flush_seconds – Number of seconds between flushing pending events to persistent storage.

Methods:

add_scalar(*, tag, value, timestep)	Add a scalar value to the summary.
close()	Close the summary writer.
flush()	Flush pending events to persistent storage.

add_scalar(*, tag: str, value: float, timestep: int) → None

Add a scalar value to the summary.

Parameters:

• tag – Name of the scalar record.

• value – Value of the scalar.

• timestep – Global step value to record.

close() → None

Close the summary writer.

flush() → None

Flush pending events to persistent storage.