Create a Frengression Model

Constructs a frengression model for cross-sectional causal data simulation. The model learns the joint distribution P(X, Y, Z) decomposed into three sub-networks: the marginal P(X,Z), the causal margin P(Y|do(X)), and the association remainder eta(X,Z).

Usage

frengression(
  x_dim,
  y_dim,
  z_dim,
  num_layer = 3,
  hidden_dim = 100,
  noise_dim = 10,
  x_binary = FALSE,
  z_binary_dims = 0,
  y_binary = FALSE
)

Arguments

x_dim

Dimension of treatment variable X.

y_dim

Dimension of outcome variable Y.

z_dim

Dimension of confounder variable Z.

num_layer

Number of layers in each sub-network (default: 3).

hidden_dim

Hidden layer width (default: 100).

noise_dim

Noise dimension for stochastic generation (default: 10).

x_binary

Logical; is X binary? (default: FALSE).

z_binary_dims

Number of leading Z dimensions that are binary (default: 0).

y_binary

Logical; is Y binary? (default: FALSE).

Value

A frengression object (S3 class) containing three sub-networks and model parameters.

Examples

# \donttest{
  model <- frengression(x_dim = 1, y_dim = 1, z_dim = 1)
#> Error: Lantern is not loaded. Please use `install_torch()` to install additional dependencies.
# }