box_embeddings.modules.volume.bessel_volume

Module Contents

eps

euler_gamma

bessel_volume_approx(box_tensor: box_embeddings.parameterizations.box_tensor.BoxTensor, volume_temperature: float = 1.0, intersection_temperature: float = 1.0, log_scale: bool = True, scale: float = 1.0) → torch.Tensor

Volume of boxes. Uses the Softplus as an approximation of Bessel funtion.

Parameters

• box_tensor – input

• volume_temperature – 1/volume_temperature is the beta parameter for the softplus

• intersection_temperature – the intersection_temperature parameter (same value used in intersection).

• log_scale – Whether the output should be in log scale or not.

• scale – scale parameter. Should be left as 1.0 (default) in most cases.

Returns

Tensor of shape (…, ) when self has shape (…, 2, num_dims)

Raises

ValueError – if scale not in (0,1] or volume_temperature is 0.

class BesselApproxVolume(log_scale: bool = True, volume_temperature: float = 1.0, intersection_temperature: float = 1.0)

Bases: box_embeddings.modules.volume._volume._Volume

Uses the Softplus as an approximation of Bessel function.

forward(self, box_tensor: box_embeddings.parameterizations.box_tensor.BoxTensor) → torch.Tensor

Soft softplus base (instead of ReLU) volume.

Parameters

box_tensor – TODO

Returns

torch.Tensor