dair_pll.wandb_manager

Interface for logging training progress to Weights and Biases.

class dair_pll.wandb_manager.WeightsAndBiasesManager(run_name, directory, project_name, resume_from_id=None)

Bases: object

Manages logging of the training process.

Given a set of scalars, videos, and meshes, writes to Weights and Biases at https://wandb.ai .

run_name: str

Display name for Weights and Biases experiment run.

directory: str

Absolute path to store metadata.

project_name: str

Unique name for W&B project, analogous to a dair_pll experiment.

resume_from_id: typing.Optional[str] = None

Allow W&B to resume a unique run ID if provided.

launch()

Launches experiment run on Weights & Biases.

Return type:

str

Returns:

The run ID of the launched run.

static log_config(config)

Log experiment hyperparameter values.

static update(epoch, scalars, videos, meshes)

Write new epoch summary to Weights and Biases.

Parameters:

• epoch (int) – Current epoch in training process

• scalars (Dict[str, float]) – Scalars to log.

• videos (Dict[str, Tuple[ndarray, int]]) – Videos to log.

• meshes (Dict[str, MeshSummary]) – Meshes to log.

Return type:

None