point_dist

iskra.geometry.distances.point_dist(x: Tensor, y: Tensor, ord: int | float | str = 2, keepdim: bool = False) → Tensor[SOURCE]

Computes the distance between batches of vectors x_i and y_i.

Unlike PyTorch’s torch.cdist, this function works with torch.func transforms and allows for other norms allowed by torch.linalg.vector_norm, at the cost of higher memory usage. torch.cdist is recommended if memory usage is important.

Parameters:

• x (Tensor[Float, [Bs, Dim]]) – Batch of Dim-dimensional vectors.

• y (Tensor[Float, [Bs, Dim]]) – Batch of Dim-dimensional vectors.

• ord (int | float | str, optional) – Order of p-norm. follows same convention as PyTorch’s vector_norm. Defaults to 2.

• keepdim (bool, optional) – Whether to keep the last dimension after reduction. Defaults to False.

Raises:

ValueError – Tensors x and y must have the same shape.

Returns:

​ (Tensor[Float, [Bs] | [Bs, 1]]) – Last dimension of output is 1 if keepdim=True, otherwise it is removed.