foxai.callbacks package

Submodules

foxai.callbacks.wandb_callback module

Callback for Weights and Biases.

class foxai.callbacks.wandb_callback.WandBCallback(wandb_logger: WandbLogger, explainers: List[ExplainerWithParams], idx_to_label: Dict[int, str], max_artifacts: int = 3)[source]

Bases: Callback

Library callback for Weights and Biases.

explain(model: LightningModule, item: Tensor, target_label: Tensor, attributes_dict: Dict[str, List[ndarray]], caption_dict: Dict[str, List[str]], figures_dict: Dict[str, List[Figure]]) → Tuple[Dict[str, List[ndarray]], Dict[str, List[str]], Dict[str, List[Figure]]][source]

Calculate explainer attributes, creates captions and figures.

Parameters:

  • model – Model to explain.

  • item – Input data sample tensor.

  • target_label – Sample label.

  • attributes_dict – List of attributes for every explainer and sample.

  • caption_dict – List of captions for every explainer and sample.

  • figures_dict – List of figures for every explainer and sample.

Returns:

Tuple of maps containing attributes, captions and figures for every explainer and sample.

iterate_dataloader(dataloader_list: List[DataLoader], max_items: int) → Generator[Tuple[Tensor, Tensor], None, None][source]

Iterate over dataloader list with constraint on max items returned.

Parameters:

  • dataloader – Trainer dataloader.

  • max_items – Max items to return.

Yields:

Tuple containing training sample and corresponding label.

log_explanations(attributes_dict: Dict[str, List[ndarray]], caption_dict: Dict[str, List[str]], figures_dict: Dict[str, List[Figure]]) → None[source]

Log explanation artifacts to W&B experiment.

Parameters:

  • attributes_dict – Numpy array attributes for every sample and every explainer.

  • caption_dict – Caption for every sample and every explainer.

  • figures_dict – Figure with attributes for every sample and every explainer.

on_train_start(trainer: Trainer, pl_module: LightningModule) → None[source]

Save index to labels mapping and validation samples to experiment at fit.

Parameters:

  • trainer – Trainer object.

  • pl_module – Model to explain.

on_validation_epoch_end(trainer: Trainer, pl_module: LightningModule) → None[source]

Export model’s state dict in log directory on validation epoch end.

Parameters:

  • trainer – Trainer object.

  • pl_module – Model to explain.

Module contents