Build SPQRX Neural Network Architecture

Constructs the Keras model architecture for the heavy-tail SPQRX (Spline-based Probabilistic Quantile Regression) model.

Usage

SPQRX(input_dim, hidden_dim, k, activation = "relu")

Arguments

input_dim

Integer. Dimension of the covariate input.

hidden_dim

Integer vector. Number of units in each hidden layer.

k

Integer. Number of spline basis functions.

activation

Provide the activation function for the hidden layers.

Value

A keras_model object representing the SPQRX architecture.

Details

The network maps covariates to spline mixture probabilities and a tail parameter \(\xi\). The model output concatenates:

• Mixture probabilities over spline basis functions

• Tail parameter \(\xi\)

• Observed response values

• Spline basis evaluations

This function defines only the network architecture. Model compilation, training, and loss specification are handled separately.