SequentialModule

Description

A function factory that creates neural network architectures for Riesz representer estimation.

Usage

sequential_module(; layers::Int=1, hidden::Int=20, dropout::Float64=0.1)

Arguments

Argument	Default	Description
layers	1	Number of hidden layers
hidden	20	Number of hidden units
dropout	0.1	Dropout rate

Value

Returns a function that takes input dimension and returns a Flux Chain.

Examples

# Default architecture
nn = sequential_module()

# Custom architecture
nn = sequential_module(layers = 2, hidden = 64, dropout = 0.2)

# Use with crumble
result = crumble(
    data = data,
    trt = ["A"],
    outcome = "Y",
    mediators = ["M"],
    covar = ["W1"],
    nn_module = nn,
    control = crumble_control(epochs = 50)
)

Architecture Details

The default architecture: - Input layer: d_in → hidden with ELU activation - Hidden layers: hidden → hidden with ELU (if layers > 0) - Output layer: hidden → 1 with Dropout and Softplus