openfold.model.dropout

Classes

Dropout(r, batch_dim)	Implementation of dropout with the ability to share the dropout mask along a particular dimension.
DropoutColumnwise(r, *[, batch_dim])	Convenience class for columnwise dropout as described in subsection 1.11.6.
DropoutRowwise(r, *[, batch_dim])	Convenience class for rowwise dropout as described in subsection 1.11.6.

class Dropout(r, batch_dim)

Bases: Module

Implementation of dropout with the ability to share the dropout mask along a particular dimension.

If not in training mode, this module computes the identity function.

Parameters:

• r (float)

• batch_dim (int | List[int])

forward(x)

Parameters:

x (Tensor) – Tensor to which dropout is applied. Can have any shape compatible with self.batch_dim

Return type:

Tensor

class DropoutColumnwise(r, *, batch_dim=-2)

Bases: Dropout

Convenience class for columnwise dropout as described in subsection 1.11.6.

Parameters:

• r (float)

• batch_dim (int | List[int])

class DropoutRowwise(r, *, batch_dim=-3)

Bases: Dropout

Convenience class for rowwise dropout as described in subsection 1.11.6.

Parameters:

• r (float)

• batch_dim (int | List[int])