XDER

Arguments

Options

--alphafloat

Help: Penalty weight.

• Default: None

--betafloat

Help: Penalty weight.

• Default: None

--simclr_tempfloat

Help: Temperature for SimCLR loss

• Default: 5

--gammafloat

Help: Weight for logit update

• Default: 0.85

--simclr_batch_sizeint

Help: Batch size for SimCLR loss

• Default: 64

--simclr_num_augint

Help: Number of augmentations for SimCLR loss

• Default: 4

--lambdfloat

Help: Weight for consistency loss

• Default: 0.05

--constr_etafloat

Help: Regularization weight for past/future constraints

• Default: 0.1

--constr_marginfloat

Help: Margin for past/future constraints

• Default: 0.3

--dp_weightfloat

Help: Weight for distance preserving loss

• Default: 0

--past_constraint0|1|True|False -> bool

Help: Enable past constraint

• Default: 1

--future_constraint0|1|True|False -> bool

Help: Enable future constraint

• Default: 1

--align_bn0|1|True|False -> bool

Help: Use BatchNorm alignment

• Default: 0

Rehearsal arguments

Arguments shared by all rehearsal-based methods.

--buffer_sizeint

Help: The size of the memory buffer.

• Default: None

--minibatch_sizeint

Help: The batch size of the memory buffer.

• Default: None

Classes

class models.xder.XDerV2(backbone, loss, args, transform, dataset=None)

Bases: ContinualModel

Continual learning via eXtended Dark Experience Replay.

COMPATIBILITY: List[str] = ['class-il', 'domain-il', 'task-il', 'general-continual']

NAME: str = 'xder'

end_task(dataset)

static get_parser(parser)

Return type:

ArgumentParser

observe(inputs, labels, not_aug_inputs, epoch=None)

update_logits(old, new, gt, task_start, n_tasks=1)