Functions — github.com/ayushkumartarun/zero-shot-unlearning

CodeHub MCPcopy Create free account

hub / github.com/ayushkumartarun/zero-shot-unlearning / functions

Functions47 in github.com/ayushkumartarun/zero-shot-unlearning

⨍ Functions 47 ◇ Types & classes 11

↓ 8 callersMethod__init__

(self, in_channels, out_channels, kernel_size=3, stride=1, padding=None, output_padding=0, ac

↓ 8 callersFunctionevaluate

(model, val_loader, device='cuda')

↓ 3 callersFunctioncollect_prob

(data_loader, model)

↓ 3 callersFunctionentropy

(p, dim = -1, keepdim = False)

↓ 2 callersFunctionattention

Taken from https://github.com/szagoruyko/attention-transfer :param x = activations

↓ 2 callersFunctiondivergence

(student_logits, teacher_logits, KL_temperature)

↓ 2 callersFunctionrelearn_time

(model, train_loader, valid_loader, reqAcc, lr)

↓ 2 callersFunctiontraining_step

(model, batch, device)

↓ 1 callersMethod__init__

(self, z_dim, num_channels = 3)

↓ 1 callersFunctionaccuracy

(outputs, labels)

↓ 1 callersFunctionattention_diff

Taken from https://github.com/szagoruyko/attention-transfer :param x = activations :param y = activations

↓ 1 callersFunctionepoch_end

(model, epoch, result)

↓ 1 callersFunctionget_lr

↓ 1 callersFunctionget_membership_attack_data

(retain_loader, forget_loader, test_loader, model)

↓ 1 callersMethodreset_parameters

↓ 1 callersFunctionvalidation_epoch_end

(model, outputs)

↓ 1 callersFunctionvalidation_step

(model, batch, device)

FunctionKT_loss_generator

(student_logits, teacher_logits, KL_temperature)

FunctionKT_loss_student

(student_logits, student_activations, teacher_logits, teacher_activations, KL_temperature = 1, AT_beta = 250)

(self, in_channels, out_channels, kernel_size=3, stride=1, padding=None, output_padding=0, w_sig =\

(self, filters_percentage=1., n_channels=3, num_classes=10, dropout=False, batch_norm=True)

(self, n_classes)

(self, in_channels, out_channels, kernel_size, padding, stride)

Infinite loader, which contains a learnable generator.

(full_model, model, gold_model, train_data, val_retain, val_forget, batch_size = 256, error

Functioncifar10

Functionfit_one_cycle

(epochs, max_lr, model, train_loader, val_loader, weight_decay=0, grad_clip=None, opt_func=

(self, x, true_labels=None)

Functionget_membership_attack_prob

(retain_loader, forget_loader, test_loader, model)

Methodprint_shape

For debugging purposes

:return: if grid returns single grid image, else returns n images.