espnet2.gan_tts.vits.transform.rational_quadratic_spline

source

espnet2.gan_tts.vits.transform.rational_quadratic_spline(inputs, unnormalized_widths, unnormalized_heights, unnormalized_derivatives, inverse=False, left=0.0, right=1.0, bottom=0.0, top=1.0, min_bin_width=0.001, min_bin_height=0.001, min_derivative=0.001)

Applies a rational quadratic spline transformation to the inputs.

This function computes the rational quadratic spline transformation for a given set of inputs based on specified widths, heights, and derivatives. The transformation can be performed in both forward and inverse directions based on the inverse parameter.

• Parameters:
  • inputs (torch.Tensor) – The input tensor to be transformed.
  • unnormalized_widths (torch.Tensor) – The unnormalized widths of the bins.
  • unnormalized_heights (torch.Tensor) – The unnormalized heights of the bins.
  • unnormalized_derivatives (torch.Tensor) – The unnormalized derivatives at the bin edges.
  • inverse (bool , optional) – If True, performs the inverse transformation. Defaults to False.
  • left (float , optional) – The left boundary of the input range. Defaults to 0.0.
  • right (float , optional) – The right boundary of the input range. Defaults to 1.0.
  • bottom (float , optional) – The bottom boundary of the output range. Defaults to 0.0.
  • top (float , optional) – The top boundary of the output range. Defaults to 1.0.
  • min_bin_width (float , optional) – The minimum allowed width for the bins. Defaults to 1e-3.
  • min_bin_height (float , optional) – The minimum allowed height for the bins. Defaults to 1e-3.
  • min_derivative (float , optional) – The minimum allowed derivative value. Defaults to 1e-3.
• Returns: A tuple containing: : - outputs (torch.Tensor): The transformed output tensor.
  • logabsdet (torch.Tensor): The log absolute determinant of the transformation.
• Return type: tuple
• Raises:ValueError – If the input is not within the specified domain or if the minimal bin width or height is too large for the number of bins.

Examples

>>> inputs = torch.tensor([0.1, 0.5, 0.9])
>>> widths = torch.tensor([[0.1, 0.2], [0.1, 0.2], [0.1, 0.2]])
>>> heights = torch.tensor([[0.3, 0.4], [0.3, 0.4], [0.3, 0.4]])
>>> derivatives = torch.tensor([[0.1, 0.1], [0.1, 0.1], [0.1, 0.1]])
>>> outputs, logabsdet = rational_quadratic_spline(inputs, widths,
... heights, derivatives)
>>> print(outputs)
>>> print(logabsdet)

NOTE

The implementation relies on the properties of rational quadratic splines, which can be beneficial in various applications such as normalizing flows in generative models.