nip.utils.torch.PairedGaussianNoise

class nip.utils.torch.PairedGaussianNoise(sigma: float, pair_dim: int = 0, train_sigma: bool = False, dtype=torch.float32)

Add Gaussian noise copied across the graph pair dimension.

Parameters:

• sigma (float) – The relative standard deviation of the Gaussian noise. This will be multiplied by the magnitude of the input to get the standard deviation for the noise.

• pair_dim (int, default=0) – The graph pair dimension.

• train_sigma (bool, default=False) – Whether the sigma parameter should be trained or not.

Notes

Adapted from https://discuss.pytorch.org/t/where-is-the-noise-layer-in-pytorch/2887/4

Methods Summary

__init__(sigma[, pair_dim, train_sigma, dtype])	Initialize internal Module state, shared by both nn.Module and ScriptModule.
__repr__()	Return repr(self).
forward(x)	Add Gaussian noise to the input tensor.
to(*args, **kwargs)	Move the module to a new device or dtype.

Attributes

T_destination
call_super_init
dump_patches
training

Methods

__init__(sigma: float, pair_dim: int = 0, train_sigma: bool = False, dtype=torch.float32)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

__repr__()

Return repr(self).

forward(x: Tensor) → Tensor

Add Gaussian noise to the input tensor.

Parameters:

x (Tensor) – The input tensor.

Returns:

x_noisy (Tensor) – The input tensor with Gaussian noise added.

to(*args, **kwargs) → PairedGaussianNoise

Move the module to a new device or dtype.

Parameters:

• *args – Arguments to pass to the to method of the superclass.

• **kwargs – Keyword arguments to pass to the to method of the superclass.

Returns:

self (PairedGaussianNoise) – The module itself.