semipy.methods.utils.pseudolabel_loss
Warning
This section is in construction.
This function computes the labelled/unlabelled loss with respect to PseudoLabel method.
Parameters
- model - The model to train.
- X - A
torch.tensor
representing one batch of data (both labelled and unlabelled). - y - A
torch.tensor
representing the labels of the current batch (unlabelled are labelled as -1). - lbda (float) - Balancing weight for unlabelled loss.
- threshold (float) - Probability threshold for pseudo-labelling.
- debiased (bool) - To activate or not safe-SSL (Schmutz et al.) via debiasing. Default:
False