XDER RPC CASPER

Arguments

Options

--alphafloat

Help: Penalty weight.

• Default: None

--betafloat

Help: Penalty weight.

• Default: None

--gammafloat

Help: None

• Default: 0.85

--etafloat

Help: None

• Default: 0.1

--mfloat

Help: None

• Default: 0.3

--clip_gradnone_or_float

Help: Clip gradient norm (default: None, no clipping)

• Default: None

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

Help: Use BatchNorm alignment

• Default: 0

--n_rpc_headsint

Help: N Heads for RPC

• Default: None

--casper_batchint

Help: Size of minibatch for casper. Equal to batch_size by default, if negative equal to buffer_size.

• Default: None

--rhofloat

Help: Weight for casper loss.

• Default: 0.01

--knn_laplaceint

Help: K of knn to build the graph for laplacian.

• Default: 10

--pint

Help: Number of classes to be drawn from the buffer. Default is N_CLASSES_PER_TASK.

• Default: None

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_rpc_casper.XDerRPCCasper(backbone, loss, args, transform, dataset=None)

Bases: CasperModel

Continual learning via eXtended Dark Experience Replay with RPC. Treated with CaSpeR!

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

NAME: str = 'xder_rpc_casper'

end_task(dataset)

forward(x)

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)

Functions

models.xder_rpc_casper.dsimplex(num_classes=10)