espnet2.gan_tts.vits.transform.unconstrained_rational_quadratic_spline

source

espnet2.gan_tts.vits.transform.unconstrained_rational_quadratic_spline(inputs, unnormalized_widths, unnormalized_heights, unnormalized_derivatives, inverse=False, tails='linear', tail_bound=1.0, min_bin_width=0.001, min_bin_height=0.001, min_derivative=0.001)

Unconstrained Rational Quadratic Spline Transformation.

This function implements an unconstrained rational quadratic spline for transformation, which can be used in various applications, such as normalizing flows. The transformation allows for flexible mapping of inputs to outputs while maintaining differentiability.

espnet2.gan_tts.vits.transform.inputs

The input tensor to be transformed.

• Type: torch.Tensor

espnet2.gan_tts.vits.transform.unnormalized_widths

The unnormalized widths for the spline bins.

• Type: torch.Tensor

espnet2.gan_tts.vits.transform.unnormalized_heights

The unnormalized heights for the spline bins.

• Type: torch.Tensor

espnet2.gan_tts.vits.transform.unnormalized_derivatives

The unnormalized derivatives at the spline knots.

• Type: torch.Tensor

espnet2.gan_tts.vits.transform.inverse

If True, performs the inverse transformation.

• Type: bool

espnet2.gan_tts.vits.transform.tails

Specifies the behavior of the tails; default is “linear”.

• Type: str

espnet2.gan_tts.vits.transform.tail_bound

The bounds for the tails of the spline.

• Type: float

espnet2.gan_tts.vits.transform.min_bin_width

The minimum allowed width of the spline bins.

• Type: float

espnet2.gan_tts.vits.transform.min_bin_height

The minimum allowed height of the spline bins.

• Type: float

espnet2.gan_tts.vits.transform.min_derivative

The minimum allowed derivative for the spline.

• Type: float

• Parameters:

  • inputs (torch.Tensor) – The input values to be transformed.
  • unnormalized_widths (torch.Tensor) – Unnormalized widths of the spline.
  • unnormalized_heights (torch.Tensor) – Unnormalized heights of the spline.
  • unnormalized_derivatives (torch.Tensor) – Unnormalized derivatives at the spline knots.
  • inverse (bool , optional) – If True, apply the inverse transformation. Defaults to False.
  • tails (str , optional) – Specifies the behavior of the tails. Defaults to “linear”.
  • tail_bound (float , optional) – Bound for the tails. Defaults to 1.0.
  • min_bin_width (float , optional) – Minimum bin width. Defaults to DEFAULT_MIN_BIN_WIDTH.
  • min_bin_height (float , optional) – Minimum bin height. Defaults to DEFAULT_MIN_BIN_HEIGHT.
  • min_derivative (float , optional) – Minimum derivative. Defaults to DEFAULT_MIN_DERIVATIVE.

• Returns: A tuple containing: : - outputs (torch.Tensor): The transformed output values.

  • logabsdet (torch.Tensor): The log absolute determinant of the Jacobian of the transformation.

• Return type: tuple

• Raises:RuntimeError – If the specified tails are not implemented.

Examples

>>> inputs = torch.tensor([0.5, 0.6])
>>> widths = torch.tensor([[0.1, 0.2], [0.1, 0.2]])
>>> heights = torch.tensor([[0.5, 0.5], [0.5, 0.5]])
>>> derivatives = torch.tensor([[0.1, 0.1], [0.1, 0.1]])
>>> outputs, logabsdet = unconstrained_rational_quadratic_spline(
...     inputs, widths, heights, derivatives
... )

NOTE

The function assumes that the input tensors are properly shaped and compatible with each other.