Python v0.4.0a0+62b766a

BatchNorm1d

class dragon.vm.torch.nn.BatchNorm1d(
  num_features,
  eps=1e-05,
  momentum=0.1,
  affine=True,
  track_running_stats=True
)[source]

Apply the batch normalization over 2d input. [Ioffe & Szegedy, 2015].

The normalization is defined as:

y=x−E[x]Var[x]+ϵ∗γ+βy = \frac{x - \mathrm{E}[x]} {\sqrt{\mathrm{Var}[x] + \epsilon}} * \gamma + \beta

The running average of statistics are calculated as:

xrunning=(1−momentum)∗xrunning+momentum∗xbatchx_{\text{running}} = (1 - \text{momentum}) * x_{\text{running}} + \text{momentum} * x_{\text{batch}}

See also

torch.nn.functional.batch_norm(…)

__init__

BatchNorm1d.__init__(
  num_features,
  eps=1e-05,
  momentum=0.1,
  affine=True,
  track_running_stats=True
)[source]

Create a BatchNorm1d module.

Parameters:
  • num_features (int) – The number of channels.
  • eps (float, optional, default=1e-5) – The value to ϵ\epsilon.
  • momentum (float, optional, default=0.1) – The value to momentum\text{momentum}.
  • affine (bool, optional, default=True) – True to apply an affine transformation.
  • track_running_stats (bool, optional, default=True) – True to using stats when switching to eval.