Performs a single optimization step. Arguments: closure (callable, optional): A closure that reevaluates the model

return **class** of backbone network (not instance). Args net_name: 'WideResNet' or network names in torchvision.models from_

get_data_loader returns torch.utils.data.DataLoader for a Dataset. All arguments are comparable with those of pytorch DataLoader. However

wrapper for cross entropy loss in pytorch. Args logits: logit values, shape=[Batch size, # of classes] targets: integer

get_dset returns class BasicDataset, containing the returns of get_data. Args is_ulb: If True, returned dataset

Get cosine scheduler (LambdaLR). if warmup is needed, set num_warmup_steps (int) > 0.

return optimizer (name) in torch.optim. If bn_wd_skip, the optimizer does not apply weight decay regularization on parameters in batch no

get_ssl_dset split training samples into labeled and unlabeled samples. The labeled data is balanced samples over classes.

get_data returns data (images) and targets (labels) shape of data: B, H, W, C shape of labels: B,

1x1 convolution

3x3 convolution with padding

Returns keys1 e.g. train,eval.

Returns keys2 e.g. loss,acc.

samples for labeled data (sampling with balanced ratio over classes)

Args alg: SSL algorithms name: name of dataset in torchvision.datasets (cifar10, cifar100, svhn, stl10) t