Simple torch.distributions wrappers for DL.

This module provides the following features.

• Easy Instanitate

  import torch
  from distrubution_extensions import NormalFactory
  
  tensor = torch.rand([256, 100, 16])
  distribution = NormalFactory()(Tensor)
  distribution.sample()  # -> Tensor[256, 100, 8]

• Easy Independence

  distribution = Normal(loc=loc, scale=scale)
  independent = distribution.independent(dim=1)

• Device Conversion

  device = torch.device("cuda:0")
  distribution = Normal(loc=loc, scale=scale)
  distribution = distribution.to(device=device)

• Slicing

  distribution = Normal(loc=loc, scale=scale)[:, 0, :]
  distribution.sample()  # -> Tensor[256, 8]

• Stop Gradient

  distribution = Normal(loc=loc, scale=scale) 
  distribution.detach()