transGFM (version 1.0.2)

transGFM_multi: Multiple source transfer learning for generalized factor models

Description

Multiple source transfer learning for generalized factor models

Usage

transGFM_multi(
  source_data_list,
  target_data,
  r,
  data_type = "count",
  method = "AD",
  lambda_seq = seq(0, 10, by = 1),
  K_cv = 3,
  sigma2 = 1,
  max_iter_cjmle = 30,
  max_iter_refine = 30,
  max_iter_nuclear = 100,
  verbose = FALSE
)

Value

List containing final estimate and intermediate results

Arguments

source_data_list

List of source data matrices (may contain missing values)

target_data

Target data matrix (complete)

r

Number of factors

data_type

Type of data: "continuous", "count", or "binary"

method

Fusion method: "AD" (Average-Debias) or "DA" (Debias-Average)

lambda_seq

Sequence of lambda values for CV

K_cv

Number of CV folds

sigma2

Variance parameter for continuous data

max_iter_cjmle

Maximum iterations for CJMLE

max_iter_refine

Maximum iterations for refinement

max_iter_nuclear

Maximum iterations for nuclear MLE

verbose

Print progress information

Examples

# \donttest{
# Generate Poisson data
set.seed(2025)

# Generate 3 source datasets (100 x 100 with different missing rates)
n1 <- 100; p1 <- 100; r <- 2
source_list <- list()
F_s <- matrix(runif(n1 * r, min = -2, max = 2), n1, r)
B_s <- matrix(runif(p1 * r, min = -2, max = 2), p1, r)
M_s <- F_s %*% t(B_s)
for (s in 1:3) {
  X_s <- matrix(rpois(n1 * p1, exp(M_s)), n1, p1)

  # Add missing values (10%, 12%, 14% for sources 1-3)
  missing_rate <- 0.1 + (s - 1) * 0.02
  n_missing <- floor(n1 * p1 * missing_rate)
  missing_idx <- sample(n1 * p1, n_missing)
  X_s[missing_idx] <- NA

  source_list[[s]] <- X_s
}

# Target data (50 x 50, complete)
n0 <- 50; p0 <- 50
M_target_true <- M_s[1:n0, 1:p0]
X_target <- matrix(rpois(n0 * p0, exp(M_target_true)), n0, p0)

# Run transGFM_multi with AD method
result_AD <- transGFM_multi(
  source_data_list = source_list,
  target_data = X_target,
  r = 2,
  data_type = "count",
  method = "AD",
  lambda_seq = seq(0, 5, by = 1),
  K_cv = 3,
  verbose = FALSE
)

# Run transGFM_multi with DA method
result_DA <- transGFM_multi(
  source_data_list = source_list,
  target_data = X_target,
  r = 2,
  data_type = "count",
  method = "DA",
  verbose = FALSE
)

# Compare results
print(paste("AD method error:", relative_error(result_AD$M_trans, M_target_true)))
print(paste("DA method error:", relative_error(result_DA$M_trans, M_target_true)))
# }