peal.generators.tabular_ddpm.BasicDiscreteTimeModel

class peal.generators.tabular_ddpm.BasicDiscreteTimeModel[source]

Bases: Module

Score network plus timestep embedding: the trainable part of the DDPM.

Parameters:

  • input_dim (int) – Feature dimension of the data.

  • embed_dim (int) – Width of the timestep embedding and MLP.

  • num_timesteps (int) – Size of the positional-encoding table.

Methods

forward

Return the predicted noise for x at timestep(s) t.
__init__(input_dim, embed_dim, num_timesteps)[source]

Instantiate the PositionalEncoding and ScoreNetwork.

Parameters:

  • input_dim (int)

  • embed_dim (int)

  • num_timesteps (int)

forward(x, t)[source]

Return the predicted noise for x at timestep(s) t.