nFunNN_CR {nFunNN}

Curve reconstruction

Description

Curve reconstruction by the trained transformed functional autoassociative neural network.

Usage

nFunNN_CR(model, X_ob, L, t_grid)

Arguments

model

The trained transformed functional autoassociative neural network obtained from nFunNNmodel.

X_ob

A matrix denoting the observed data from subjects that we aim to predict.

L

An integer denoting the number of B-spline basis functions for the parameters in the network.

t_grid

A vector denoting the observation time grids on [0, 1].

Value

A torch tensor denoting the predicted values.

Examples

n <- 2000
m <- 51
t_grid <- seq(0, 1, length.out = m)
m_est <- 101
t_grid_est <- seq(0, 1, length.out = m_est)
err_sd <- 0.1
Z_1a <- stats::rnorm(n, 0, 3)
Z_2a <- stats::rnorm(n, 0, 2)
Z_a <- cbind(Z_1a, Z_2a)
Phi <- cbind(sin(2 * pi * t_grid), cos(2 * pi * t_grid))
Phi_est <- cbind(sin(2 * pi * t_grid_est), cos(2 * pi * t_grid_est))
X <- Z_a %*% t(Phi)
X_to_est <- Z_a %*% t(Phi_est)
X_ob <- X + matrix(stats::rnorm(n * m, 0, err_sd), nr = n, nc = m)
L_smooth <- 10
L <- 10
J <- 20
K <- 2
R <- 20
nFunNN_res <- nFunNNmodel(X_ob, t_grid, t_grid_est, L_smooth,
L, J, K, R, lr = 0.001, n_epoch = 1500, batch_size = 100)
model <- nFunNN_res$model
X_pre <- nFunNN_CR(model, X_ob, L, t_grid)
sqrt(torch::nnf_mse_loss(X_pre, torch::torch_tensor(X_to_est))$item())