Training data shuffle in Naive or other strategies?

[bartosz-grabowski]

Hi,

Is there data shuffling within each experience (not the experiences themselves) implemented for various strategies, particularly for avalanche.training.supervised.Naive strategy? I tried to find it but was not able to confirm whether the shuffling is being carried out during training.

[AntonioCarta]

Yes, there is and it's done by the Dataloader. By default, Avalanche uses a TaskBalancedDataloader, which shuffles and balance mini-batches by task. You can override it by setting strategy.dataloader with a plugin before_training_exp.

[bartosz-grabowski]

Thank you for your help.