chainer.functions.gaussian(mean, ln_var, *, eps=None, return_eps=False)[source]

Gaussian sampling function.

This function takes a mean \(\mu\) and the logarithm of a variance \(\log(\sigma^2)\) as inputs and outputs a sample drawn from a Gaussian distribution \(N(\mu, \sigma)\).

The inputs must have the same shape.

  • mean (Variable or N-dimensional array) – Input variable representing the mean \(\mu\).
  • ln_var (Variable or N-dimensional array) – Input variable representing the logarithm of a variance \(\log(\sigma^2)\).
  • eps (N-dimensional array or None) – The eps value to be used. You do not have to specify this value, unless you need to make results deterministic. If eps is not specified or set to None, an eps value will be generated randomly. The shape and dtype must be the same as ln_var and should be on the same device.
  • return_eps (bool) – If True, the eps value used in this function is returned together with the output variable. The returned eps can later be reused by passing it to the eps argument.

When return_eps is False (default), returns the output variable with the shape of mean and/or ln_var. When True, returns the tuple of the output variable and eps (N-dimensional array). The eps will be on the same device as the input (ln_var).

Return type:

Variable or tuple