Dropout#

class myoverse.transforms.Dropout(p=0.1, dim=None, **kwargs)[source]#

Randomly zero out elements.

Parameters:

  • p (float) – Probability of zeroing each element.

  • dim (str) – If specified, drops entire slices along this dimension. If None, drops individual elements.

Examples

>>> x = torch.randn(64, 2048, device='cuda', names=('channel', 'time'))
>>> # Element-wise dropout
>>> dropout = Dropout(p=0.1)
>>> # Channel dropout (drop entire channels)
>>> channel_dropout = Dropout(p=0.1, dim='channel')

Methods

__init__([p, dim])

_apply(x)

Apply the transform.

eval()

Set to evaluation mode.

train()

Set to training mode.
eval()[source]#

Set to evaluation mode.

train()[source]#

Set to training mode.

property training_mode: bool#

Check if in training mode (dropout only during training).