dagma.nonlinear.DagmaMLP¶

class dagma.nonlinear.DagmaMLP(dims: list[int], bias: bool = True, dtype: torch.dtype = torch.double)¶

Bases: torch.nn.Module

Class that models the structural equations for the causal graph using MLPs.

Parameters:¶

dims : List[int]¶

Number of neurons in hidden layers of each MLP representing each structural equation.

bias : bool, optional¶

Flag whether to consider bias or not, by default True

dtype : torch.dtype, optional¶

Float precision, by default torch.double

Methods¶

forward(→ torch.Tensor)	Applies the current states of the structural equations to the dataset X
h_func(→ torch.Tensor)	Constrain 2-norm-squared of fc1 weights along m1 dim to be a DAG
fc1_l1_reg(→ torch.Tensor)	Takes L1 norm of the weights in the first fully-connected layer
fc1_to_adj(→ numpy.ndarray)	Computes the induced weighted adjacency matrix W from the first FC weights.