starling.data.schedulers.sigmoid_beta_schedule

sigmoid_beta_schedule(timesteps: int, start: int = 3, end: int = 3, tau: int = 1) Tensor[source]

The beta values are generated using a sigmoid function. The beta values are clipped between 0 and 0.999.

Parameters:
  • timesteps (int) – The number of timesteps which will be used to generate the beta values.

  • start (int, optional) – The starting value for the sigmoid function, by default 3

  • end (int, optional) – The ending value for the sigmoid function, by default 3

  • tau (int, optional) – The time constant for the sigmoid function, by default 1

Returns:

A tensor containing the beta values.

Return type:

torch.Tensor