From 1bcb878be8f9926e00a3cb10450d3899ebccfda3 Mon Sep 17 00:00:00 2001 From: Chao Cheng Date: Wed, 21 Aug 2024 13:05:32 +0800 Subject: [PATCH] Update documentation of LogisticFAR MM related solvers --- R/mm_path_solver.R | 50 +++++++++++++++++++------------ man/Logistic_FARMM_CV_path.Rd | 9 ++++-- man/Logistic_FARMM_CV_path_par.Rd | 9 ++++-- man/Logistic_FARMM_Path.Rd | 9 ++++-- 4 files changed, 49 insertions(+), 28 deletions(-) diff --git a/R/mm_path_solver.R b/R/mm_path_solver.R index b122bd8..db40831 100644 --- a/R/mm_path_solver.R +++ b/R/mm_path_solver.R @@ -12,8 +12,11 @@ #' First h columns are for demographical covariates(can include an intercept term) #' Rest columns are for p functional covariates, each being represented by a set of basis functions resulting kn covariates. #' -#' @param subj_vec vector of subject ID (can be integer, character or factor type), -#' used for mixture model post-selection estimation +#' @param rand_eff_df `data.frame` of random effect related data. It must contain +#' at least one column named "subj_vec_fct", which indicates the subject level. +#' If this is the only column in `rand_eff_df`, then a constant random effect +#' is applied. If there is other column(s), then they will all be additively added +#' to the random effect as the slope term. #' #' @param h,kn,p dimension information for the dataset(\code{x_mat}). #' @@ -67,7 +70,7 @@ #' @return A list containing the solution path of \code{delta}, \code{eta_stack}, \code{mu1} #' and some computation information such as convergency, iteration number and the lambda #' sequence of this solution path. -Logistic_FARMM_Path <- function(y_vec, x_mat, h, kn, p, subj_vec, +Logistic_FARMM_Path <- function(y_vec, x_mat, h, kn, p, rand_eff_df, p_type, p_param, lambda_seq, lambda_length, min_lambda_ratio = 0.01, mu2, a = 1, bj_vec = 1, cj_vec = sqrt(kn), rj_vec = 10^(-6), @@ -402,8 +405,11 @@ Logistic_FARMM_Path <- function(y_vec, x_mat, h, kn, p, subj_vec, #' First h columns are for demographical covariates(can include an intercept term) #' Rest columns are for p functional covariates, each being represented by a set of basis functions resulting kn covariates. #' -#' @param subj_vec vector of subject ID (can be integer, character or factor type), -#' used for mixture model post-selection estimation +#' @param rand_eff_df `data.frame` of random effect related data. It must contain +#' at least one column named "subj_vec_fct", which indicates the subject level. +#' If this is the only column in `rand_eff_df`, then a constant random effect +#' is applied. If there is other column(s), then they will all be additively added +#' to the random effect as the slope term. #' #' @param h,kn,p dimension information for the dataset(\code{x_mat}). #' @@ -477,7 +483,7 @@ Logistic_FARMM_Path <- function(y_vec, x_mat, h, kn, p, subj_vec, #' averaged loglik on the testsets. It is more recommended to use the stand alone #' \code{*_pick} functions in this packages, such as \code{CV_Pick} to find a optimal #' lambda since those functions give more flexibility. -Logistic_FARMM_CV_path <- function(y_vec, x_mat, h, kn, p, subj_vec, +Logistic_FARMM_CV_path <- function(y_vec, x_mat, h, kn, p, rand_eff_df, p_type, p_param, lambda_seq, lambda_length, min_lambda_ratio = 0.01, mu2, a = 1, bj_vec = rep(1 / sqrt(kn), p), cj_vec = rep(1, p), rj_vec = 0.00001, @@ -749,7 +755,7 @@ Logistic_FARMM_CV_path <- function(y_vec, x_mat, h, kn, p, subj_vec, # find solution path on the training set print(paste("Find solution path on training set...")) train_res <- Logistic_FARMM_Path(y_vec = y_vec_train, x_mat = x_mat_train, - h = h, kn = kn, p = p, subj_vec = subj_vec, + h = h, kn = kn, p = p, rand_eff_df = rand_eff_df, p_type = p_type, p_param = p_param, lambda_seq = lambda_seq, mu2 = mu2, a = a, bj_vec = bj_vec, cj_vec = cj_vec, rj_vec = rj_vec, @@ -772,7 +778,7 @@ Logistic_FARMM_CV_path <- function(y_vec, x_mat, h, kn, p, subj_vec, # post_res <- train_res for(lam_id in 1 : lambda_length){ post_est <- Logistic_FARMM_Path_Further_Improve( - x_mat = x_mat_train, y_vec = y_vec_train, subj_vec = subj_vec, + x_mat = x_mat_train, y_vec = y_vec_train, rand_eff_df = rand_eff_df, h = h, k_n = kn, p = p, delta_vec_init = train_res$delta_path[lam_id, ], eta_stack_init = train_res$eta_stack_path[lam_id, ], @@ -804,7 +810,7 @@ Logistic_FARMM_CV_path <- function(y_vec, x_mat, h, kn, p, subj_vec, # find the lambda with the highest test loglik lam_id <- which.max(colSums(loglik_test_mat)) res <- Logistic_FARMM_Path(y_vec = y_vec, x_mat = x_mat_bak, - h = h, kn = kn, p = p, subj_vec = subj_vec, + h = h, kn = kn, p = p, rand_eff_df = rand_eff_df, p_type = p_type, p_param = p_param, lambda_seq = lambda_seq, mu2 = mu2, a = a, bj_vec = bj_vec, cj_vec = cj_vec, rj_vec = rj_vec, @@ -821,7 +827,7 @@ Logistic_FARMM_CV_path <- function(y_vec, x_mat, h, kn, p, subj_vec, if(post_selection){ lam_post_id <- which.max(colSums(loglik_post_mat)) post_est <- Logistic_FARMM_Path_Further_Improve( - x_mat = x_mat_bak, y_vec = y_vec, subj_vec = subj_vec, + x_mat = x_mat_bak, y_vec = y_vec, rand_eff_df = rand_eff_df, h = h, k_n = kn, p = p, delta_vec_init = res$delta_path[lam_post_id, ], eta_stack_init = res$eta_stack_path[lam_post_id, ], @@ -854,8 +860,11 @@ Logistic_FARMM_CV_path <- function(y_vec, x_mat, h, kn, p, subj_vec, #' First h columns are for demographical covariates(can include an intercept term) #' Rest columns are for p functional covariates, each being represented by a set of basis functions resulting kn covariates. #' -#' @param subj_vec vector of subject ID (can be integer, character or factor type), -#' used for mixture model post-selection estimation +#' @param rand_eff_df `data.frame` of random effect related data. It must contain +#' at least one column named "subj_vec_fct", which indicates the subject level. +#' If this is the only column in `rand_eff_df`, then a constant random effect +#' is applied. If there is other column(s), then they will all be additively added +#' to the random effect as the slope term. #' #' @param h,kn,p dimension information for the dataset(\code{x_mat}). #' @@ -929,7 +938,7 @@ Logistic_FARMM_CV_path <- function(y_vec, x_mat, h, kn, p, subj_vec, #' averaged loglik on the testsets. It is more recommended to use the stand alone #' \code{*_pick} functions in this packages, such as \code{CV_Pick} to find a optimal #' lambda since those functions give more flexibility. -Logistic_FARMM_CV_path_par <- function(y_vec, x_mat, h, kn, p, subj_vec, +Logistic_FARMM_CV_path_par <- function(y_vec, x_mat, h, kn, p, rand_eff_df, p_type, p_param, lambda_seq, lambda_length, min_lambda_ratio = 0.01, mu2, a = 1, bj_vec = rep(1 / sqrt(kn), p), cj_vec = rep(1, p), rj_vec = 0.00001, @@ -1186,7 +1195,7 @@ Logistic_FARMM_CV_path_par <- function(y_vec, x_mat, h, kn, p, subj_vec, } pb <- progressr::progressor(along = 1 : (nfold + 1)) # including the final estimation - cv_res <- future.apply::future_lapply(1 : nfold, function(cv_id, x_mat, y_vec, h, kn, p, subj_vec, + cv_res <- future.apply::future_lapply(1 : nfold, function(cv_id, x_mat, y_vec, h, kn, p, rand_eff_df, p_type, p_param, lambda_seq, mu2, a, bj_vec, cj_vec, rj_vec, weight_vec, logit_weight_vec, weight_already_combine, @@ -1204,11 +1213,14 @@ Logistic_FARMM_CV_path_par <- function(y_vec, x_mat, h, kn, p, subj_vec, weight_vec_test <- weight_vec[test_id_vec] logit_weight_vec_test <- logit_weight_vec[test_id_vec] + rand_eff_df_train <- rand_eff_df[-test_id_vec, , drop = FALSE] + rand_eff_df_test <- rand_eff_df[test_id_vec, , drop = FALSE] + # find solution path on the training set print(paste("Find solution path on training set...")) train_res <- Logistic_FARMM_Path( y_vec = y_vec_train, x_mat = x_mat_train, - h = h, kn = kn, p = p, subj_vec = subj_vec, + h = h, kn = kn, p = p, rand_eff_df = rand_eff_df_train, p_type = p_type, p_param = p_param, lambda_seq = lambda_seq, mu2 = mu2, a = a, bj_vec = bj_vec, cj_vec = cj_vec, rj_vec = rj_vec, @@ -1233,7 +1245,7 @@ Logistic_FARMM_CV_path_par <- function(y_vec, x_mat, h, kn, p, subj_vec, # post_res <- train_res for(lam_id in 1 : lambda_length){ post_est <- Logistic_FARMM_Path_Further_Improve( - x_mat = x_mat_train, y_vec = y_vec_train, subj_vec = subj_vec, + x_mat = x_mat_train, y_vec = y_vec_train, rand_eff_df = rand_eff_df_train, h = h, k_n = kn, p = p, delta_vec_init = train_res$delta_path[lam_id, ], eta_stack_init = train_res$eta_stack_path[lam_id, ], @@ -1258,7 +1270,7 @@ Logistic_FARMM_CV_path_par <- function(y_vec, x_mat, h, kn, p, subj_vec, return(loglik_test_mat) - }, x_mat = x_mat_bak, y_vec = y_vec, h = h, kn = kn, p = p, subj_vec = subj_vec, + }, x_mat = x_mat_bak, y_vec = y_vec, h = h, kn = kn, p = p, rand_eff_df = rand_eff_df, p_type = p_type, p_param = p_param, lambda_seq = lambda_seq, mu2 = mu2, a = a, bj_vec = bj_vec, cj_vec = cj_vec, rj_vec = rj_vec, weight_vec = weight_vec, logit_weight_vec = logit_weight_vec, weight_already_combine = weight_already_combine, @@ -1277,7 +1289,7 @@ Logistic_FARMM_CV_path_par <- function(y_vec, x_mat, h, kn, p, subj_vec, # find the lambda with the highest test loglik lam_id <- which.max(colSums(loglik_test_mat)) res <- Logistic_FARMM_Path(y_vec = y_vec, x_mat = x_mat_bak, - h = h, kn = kn, p = p, subj_vec = subj_vec, + h = h, kn = kn, p = p, rand_eff_df = rand_eff_df, p_type = p_type, p_param = p_param, lambda_seq = lambda_seq, mu2 = mu2, a = a, bj_vec = bj_vec, cj_vec = cj_vec, rj_vec = rj_vec, @@ -1296,7 +1308,7 @@ Logistic_FARMM_CV_path_par <- function(y_vec, x_mat, h, kn, p, subj_vec, # print(paste("lam_post_id = ", lam_post_id, sep = "")) # print(paste("delta_vec = ", res$delta_path[lam_post_id, ], sep = "")) post_est <- Logistic_FARMM_Path_Further_Improve( - x_mat = x_mat_bak, y_vec = y_vec, subj_vec = subj_vec, + x_mat = x_mat_bak, y_vec = y_vec, rand_eff_df = rand_eff_df, h = h, k_n = kn, p = p, delta_vec_init = res$delta_path[lam_post_id, ], eta_stack_init = res$eta_stack_path[lam_post_id, ], diff --git a/man/Logistic_FARMM_CV_path.Rd b/man/Logistic_FARMM_CV_path.Rd index 38caadc..60c5d7f 100644 --- a/man/Logistic_FARMM_CV_path.Rd +++ b/man/Logistic_FARMM_CV_path.Rd @@ -10,7 +10,7 @@ Logistic_FARMM_CV_path( h, kn, p, - subj_vec, + rand_eff_df, p_type, p_param, lambda_seq, @@ -48,8 +48,11 @@ Rest columns are for p functional covariates, each being represented by a set of \item{h, kn, p}{dimension information for the dataset(\code{x_mat}).} -\item{subj_vec}{vector of subject ID (can be integer, character or factor type), -used for mixture model post-selection estimation} +\item{rand_eff_df}{\code{data.frame} of random effect related data. It must contain +at least one column named "subj_vec_fct", which indicates the subject level. +If this is the only column in \code{rand_eff_df}, then a constant random effect +is applied. If there is other column(s), then they will all be additively added +to the random effect as the slope term.} \item{p_type}{an character variable indicating different types of the penalty} diff --git a/man/Logistic_FARMM_CV_path_par.Rd b/man/Logistic_FARMM_CV_path_par.Rd index 6af165e..bacaa4a 100644 --- a/man/Logistic_FARMM_CV_path_par.Rd +++ b/man/Logistic_FARMM_CV_path_par.Rd @@ -10,7 +10,7 @@ Logistic_FARMM_CV_path_par( h, kn, p, - subj_vec, + rand_eff_df, p_type, p_param, lambda_seq, @@ -48,8 +48,11 @@ Rest columns are for p functional covariates, each being represented by a set of \item{h, kn, p}{dimension information for the dataset(\code{x_mat}).} -\item{subj_vec}{vector of subject ID (can be integer, character or factor type), -used for mixture model post-selection estimation} +\item{rand_eff_df}{\code{data.frame} of random effect related data. It must contain +at least one column named "subj_vec_fct", which indicates the subject level. +If this is the only column in \code{rand_eff_df}, then a constant random effect +is applied. If there is other column(s), then they will all be additively added +to the random effect as the slope term.} \item{p_type}{an character variable indicating different types of the penalty} diff --git a/man/Logistic_FARMM_Path.Rd b/man/Logistic_FARMM_Path.Rd index aa2bbc1..7e42f76 100644 --- a/man/Logistic_FARMM_Path.Rd +++ b/man/Logistic_FARMM_Path.Rd @@ -10,7 +10,7 @@ Logistic_FARMM_Path( h, kn, p, - subj_vec, + rand_eff_df, p_type, p_param, lambda_seq, @@ -46,8 +46,11 @@ Rest columns are for p functional covariates, each being represented by a set of \item{h, kn, p}{dimension information for the dataset(\code{x_mat}).} -\item{subj_vec}{vector of subject ID (can be integer, character or factor type), -used for mixture model post-selection estimation} +\item{rand_eff_df}{\code{data.frame} of random effect related data. It must contain +at least one column named "subj_vec_fct", which indicates the subject level. +If this is the only column in \code{rand_eff_df}, then a constant random effect +is applied. If there is other column(s), then they will all be additively added +to the random effect as the slope term.} \item{p_type}{an character variable indicating different types of the penalty}