avalanche.benchmarks.class_incremental_benchmark

avalanche.benchmarks.class_incremental_benchmark(datasets_dict: Dict[str, ClassificationDataset], *, class_order: Sequence[int] | None = None, num_experiences: int | None = None, num_classes_per_exp: Sequence[int] | None = None, seed: int | None = None) → CLScenario[source]

Splits datasets according to a class-incremental scenario.

Each dataset will create a stream with the same class order.

Parameters:

  • datasets_dict – A dictionary with stream names as keys (str) and AvalancheDataset as values. Usually, you want to provide at least train and test stream.

  • class_order – List of classes that determine the order of appearance in the stream. If None, random classes will be used. Defaults to None (random classes).

  • num_experiences – desired number of experiences in the stream.

  • num_classes_per_exp – If not None, a list with the number of classes to pick for each experience.

  • seed – The seed to use for random shuffling if class_order is None. If None, the current PyTorch random number generator state will be used. Defaults to None.

Returns:

A class-incremental CLScenario.