Help for package IDLFM

Title:

Individual Dynamic Latent Factor Model

Version:

1.0.0

Maintainer:

Siyang Liu <liusiyang.lucia@gmail.com>

Description:

A personalized dynamic latent factor model (Zhang et al. (2024) <doi:10.1093/biomet/asae015>) for irregular multi-resolution time series data, to interpolate unsampled measurements from low-resolution time series.

Depends:

R (≥ 4.4.0)

Imports:

stats, methods, splines, SparseArray

License:

MIT + file LICENSE

Encoding:

UTF-8

RoxygenNote:

7.3.2

Suggests:

testthat (≥ 3.0.0)

Config/testthat/edition:

NeedsCompilation:

Packaged:

2025-05-11 09:37:26 UTC; 24952

Author:

Siyang Liu [aut, cre], Jiuchen Zhang [aut], Annie Qu [aut]

Repository:

CRAN

Date/Publication:

2025-05-11 09:50:02 UTC

Individualized Dynamic Latent Factor Model

Description

This function implements the individualized dynamic latent factor model.

Usage

IDLFM(
  X,
  Y,
  n_patients,
  n_var,
  time,
  idx_x,
  idx_y,
  rank,
  k,
  N,
  lambda1 = 1,
  lambda2 = 1,
  Niter = 100,
  alpha = 0.001,
  ebs = 1e-04,
  l = 1,
  verbose
)

Arguments

X

a sparse matrix for predictor variables

Y

a sparse matrix for response variables

n_patients

the number of patients

n_var

the number of X variables

time

maximum time

idx_x

indices for the X data, a sparse matrix

idx_y

indices for the Y data, a sparse matrix

rank

rank for the random matrices

k

spline smoothness

N

number of knots in the spline

lambda1

regularization parameter for fused lasso, with the default value 1

lambda2

regularization parameter for total variation, with the default value 1

Niter

number of iterations for the Adam optimizer, with the default value 100

alpha

learning rate for the Adam optimizer, with the default value 0.001

ebs

convergence threshold, with the default value 0.0001

l

regularization parameter, with the default value 1

verbose

to control the console output

Value

A list is returned, containing the model weights, factor matrix, spline knots, predicted X and Y.

References

Zhang, J., F. Xue, Q. Xu, J. Lee, and A. Qu. "Individualized dynamic latent factor model for multi-resolutional data with application to mobile health." Biometrika (2024): asae015.

Examples

library(splines)
#if (!require("BiocManager", quietly = TRUE))
#install.packages("BiocManager")
#BiocManager::install("SparseArray")
library(SparseArray)

I <- 3
J <- 5
time <- 1000
R <- 3
k <- 3
N <- 300
idx_x <- randomSparseArray(c(I, J, time), density=0.8)
idx_y_train <- randomSparseArray(c(I, 1, time), density=0.2)
idx_y_test <- randomSparseArray(c(I, 1, time), density=0.2)
data <- generate_data(I, J, time, idx_x, idx_y_train, R, k, N)
output_x <- data[[1]]
output_y <- data[[2]]
knots <- data[[3]]
weights <- data[[4]]
Fx <- data[[5]]
Fy <- data[[6]]
IDLFM(X = output_x, Y = output_y, n_patients = I, n_var = J, time = time,
idx_x = idx_x, idx_y = idx_y_train, rank = R, k = k, N = N, verbose = FALSE)

Generate data for simulation