Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Added namespace prefixing #3

Merged
merged 2 commits into from
Dec 1, 2023
Merged

Added namespace prefixing #3

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
596ca3c
Added namespace prefixing
certara-mtalley Dec 1, 2023
a94d41b
Added ggplot2 namespace
certara-mtalley Dec 1, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 3 additions & 11 deletions DESCRIPTION
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,21 +25,13 @@ Depends:
caret,
SHAPforxgboost
Imports:
randomForest,
Boruta,
dplyr,
xgboost,
ggplot2,
glmnet,
gridExtra,
Metrics,
hrbrthemes,
Ckmeans.1d.dp,
BBmisc,
ggplot2,
GGally,
grid,
ggstance,
ggpubr,
gridExtra
xgboost
Suggests:
knitr,
rmarkdown,
Expand Down
17 changes: 0 additions & 17 deletions NAMESPACE
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,20 +2,3 @@

export(MLCovSearch)
export(generate_residualsplots)
import(BBmisc)
import(Boruta)
import(Ckmeans.1d.dp)
import(GGally)
import(Metrics)
import(SHAPforxgboost)
import(caret)
import(dplyr)
import(ggplot2)
import(ggpubr)
import(ggstance)
import(glmnet)
import(grid)
import(gridExtra)
import(hrbrthemes)
import(randomForest)
import(xgboost)
132 changes: 56 additions & 76 deletions R/cov_search.R
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,22 +1,4 @@
#' @import randomForest
#' @import Boruta
#' @import dplyr
#' @import xgboost
#' @import caret
#' @import glmnet
#' @import Metrics
#' @import hrbrthemes
#' @import Ckmeans.1d.dp
#' @import SHAPforxgboost
#' @import BBmisc
#' @import ggplot2
#' @import GGally
#' @import grid
#' @import ggstance
#' @import ggpubr
#' @import gridExtra
#'
NULL
`%>%` <- dplyr::`%>%`

#' MLCovSearch
#'
Expand All @@ -37,16 +19,16 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =
stopifnot(requireNamespace("caret", quietly = TRUE))
# Selection of columns required
tab <- tab %>%
select(ID, all_of(list_pop_param), all_of(cov_continuous), all_of(cov_factors))
dplyr::select(ID, dplyr::all_of(list_pop_param), dplyr::all_of(cov_continuous), dplyr::all_of(cov_factors))

# In order to have the individual parameter and one point per subject
dat <- unique(tab) %>%
mutate(across(all_of(cov_factors), as.factor))
dplyr::mutate(dplyr::across(dplyr::all_of(cov_factors), as.factor))

# Data for XGBoost
pop_param <- dat %>% select(all_of(list_pop_param))
factors <- dat %>% select(all_of(cov_factors))
continuous <- dat %>% select(all_of(cov_continuous))
pop_param <- dat %>% dplyr::select(dplyr::all_of(list_pop_param))
factors <- dat %>% dplyr::select(dplyr::all_of(cov_factors))
continuous <- dat %>% dplyr::select(dplyr::all_of(cov_continuous))

# One-hot encoding of categorical covariates for covariates with more than 2 levels
modified_columns <- data.frame(matrix(ncol = 0, nrow = nrow(factors)))
Expand Down Expand Up @@ -85,7 +67,7 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =
# Cross-validation
## create 5 partition of the data ( using K-1 folds (80%) as the training set and the remaining one fold (20%) as the test set repeating steps for K iterations )
x <- as.data.frame(x_xgb)
folds <- createFolds(seq(1, nrow(x_xgb)), k = 5, list = TRUE, returnTrain = FALSE)
folds <- caret::createFolds(seq(1, nrow(x_xgb)), k = 5, list = TRUE, returnTrain = FALSE)

for (j in 1:5) {
train.ind <- folds[[j]]
Expand All @@ -102,7 +84,7 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =
X <- as.matrix(training)
Y <- as.matrix(y_xgb_train)
# Perform k-fold cross-validation to find optimal lambda value
cvfit <- cv.glmnet(X, Y, alpha = 1, family = "gaussian")
cvfit <- glmnet::cv.glmnet(X, Y, alpha = 1, family = "gaussian")
# Extract the non-zero coefficients from the model at the optimal value of the regularization parameter
lasso.coef <- coef(cvfit, s = cvfit$lambda.min)[-1, ]
selected.vars <- names(lasso.coef[lasso.coef != 0])
Expand All @@ -114,19 +96,19 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =

# Boruta performed on the covariates selected by lasso
if (length(selected.vars) != 0) {
xgb.boruta <- Boruta(
xgb.boruta <- Boruta::Boruta(
train.lasso,
y = y_xgb_train,
maxRuns = 200,
doTrace = 0,
seed = 42,
getImp = getImpXgboost,
getImp = Boruta::getImpXgboost,
nrounds = 200,
objective = "reg:squarederror"
)

# Extracting the result of Boruta algorithm (keep confirmed)
boruta.df <- attStats(xgb.boruta)
boruta.df <- Boruta::attStats(xgb.boruta)
feature.imp <- row.names(boruta.df)[which(boruta.df$decision == "Confirmed")]

result_5folds[i, j] <- paste(feature.imp, collapse = ', ')
Expand All @@ -141,14 +123,14 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =
rownames(result_ML) <- list_pop_param

res <- t(result_5folds[,1:5])
res <- res %>% na_if("")
res <- res %>% dplyr::na_if("")

for (i in list_pop_param) {
list_cov <- strsplit(res[, i], ",")
list_cov_nb <- trimws(unlist(list_cov))
comptage <- as.data.frame(table(list_cov_nb))
if (nrow(comptage) != 0) {
filtered_vars <- comptage %>% filter(Freq >= 2) %>% select(list_cov_nb)
filtered_vars <- comptage %>% dplyr::filter(Freq >= 2) %>% dplyr::select(list_cov_nb)
variable_list <- as.character(filtered_vars$list_cov_nb)
cov_selected <- paste(variable_list, collapse = ", ")
result_ML[i, 1] <- cov_selected
Expand All @@ -171,8 +153,8 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =

if (is.na(result_ML[i, 1]) == FALSE){
list_cov <- strsplit(gsub(" ", "", result_ML[i, 1]), ",")
x.selected_final <- as.matrix(dat_XGB %>% select(all_of(list_cov[[1]])))
folds <- createFolds(seq(1,nrow(x.selected_final)), k = 5, list = TRUE, returnTrain = FALSE)
x.selected_final <- as.matrix(dat_XGB %>% dplyr::select(dplyr::all_of(list_cov[[1]])))
folds <- caret::createFolds(seq(1,nrow(x.selected_final)), k = 5, list = TRUE, returnTrain = FALSE)

for (j in 1:5){

Expand All @@ -188,7 +170,7 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =


if (length(list_cov[[1]]) != 0 ) {
xgb.mod <- xgboost(
xgb.mod <- xgboost::xgboost(
data = training,
label = y.xgm_train,
nrounds = 200,
Expand All @@ -199,15 +181,15 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =
# predict on the test set with the new model
y.xgb.pred <- predict(xgb.mod, newdata = testing)
# evaluate the performance of the model
RMSE[j] <- rmse(y.xgm_test,y.xgb.pred)
RMSE[j] <- Metrics::rmse(y.xgm_test,y.xgb.pred)
result_ML[i,2] <- mean(RMSE,na.rm = TRUE)



# Calculate the reference RMSE (baseline model without any covariates) by using the mean of the training y values and comparing it with the test y values
mean_y <- mean(y.xgm_train)
y.mean <- rep(mean_y, length(y.xgm_test))
RMSE_ref[j] <- rmse(y.xgm_test,y.mean)
RMSE_ref[j] <- Metrics::rmse(y.xgm_test,y.mean)
result_ML[i,3] <- mean(RMSE_ref,na.rm = TRUE)


Expand All @@ -225,7 +207,7 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =
shap_values <- SHAPforxgboost::shap.values(xgb_model = xgb_model, X_train = X_train)
shap_long <- SHAPforxgboost::shap.prep(xgb_model = xgb_model, X_train = X_train)
p <- SHAPforxgboost::shap.plot.summary(shap_long)
p <- p + ggtitle(param_name)
p <- p + ggplot2::ggtitle(param_name)

return(p)
}
Expand All @@ -236,10 +218,10 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =

if (is.na(result_ML[i, 1]) == FALSE){
list_cov <- strsplit(gsub(" ", "", result_ML[i, 1]), ",")
x.selected_final <- as.matrix(dat_XGB %>% select(all_of(list_cov[[1]])))
x.selected_final <- as.matrix(dat_XGB %>% dplyr::select(dplyr::all_of(list_cov[[1]])))

if (length(list_cov[[1]]) != 0 ) {
xgb.mod_final <- xgboost(
xgb.mod_final <- xgboost::xgboost(
data = x.selected_final,
label = y_xgb,
nrounds = 200,
Expand All @@ -258,7 +240,7 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =
}
}

combined_plots <- marrangeGrob(grobs = shap_plots,nrow = length(shap_plots),ncol = 1)
combined_plots <- gridExtra::marrangeGrob(grobs = shap_plots,nrow = length(shap_plots),ncol = 1)



Expand Down Expand Up @@ -307,24 +289,24 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac

# Selection of columns required
tab <- tab %>%
select(ID, all_of(list_pop_param), all_of(cov_continuous), all_of(cov_factors))
dplyr::select(ID, dplyr::all_of(list_pop_param), dplyr::all_of(cov_continuous), dplyr::all_of(cov_factors))



# In order to have the individual parameter and one point per subject
dat <- unique(tab) %>%
mutate(across(all_of(cov_factors), as.factor))
dplyr::mutate(dplyr::across(dplyr::all_of(cov_factors), as.factor))

# Data for XGBoost
pop_param <- dat %>% select(all_of(list_pop_param))
factors <- dat %>% select(all_of(cov_factors))
continuous <- dat %>% select(all_of(cov_continuous))
pop_param <- dat %>% dplyr::select(dplyr::all_of(list_pop_param))
factors <- dat %>% dplyr::select(dplyr::all_of(cov_factors))
continuous <- dat %>% dplyr::select(dplyr::all_of(cov_continuous))

# One-hot encoding of categorical covariates for covariates with more than 2 levels
modified_columns <- data.frame(matrix(ncol = 0, nrow = nrow(factors)))
for (col in names(factors)) {
if (is.factor(factors[[col]]) && nlevels(factors[[col]]) > 2) {
dmy <- dummyVars(paste0("~", col), data = factors)
dmy <- caret::dummyVars(paste0("~", col), data = factors)
encoded <- data.frame(predict(dmy, newdata = factors))
modified_columns <- cbind(modified_columns,encoded)
} else {
Expand All @@ -340,26 +322,24 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac
full_covariate <- c(cov_continuous, cov_factors)

res <- t(result_5folds[,1:5])
res <- res %>% na_if("")
res <- res %>% dplyr::na_if("")

result_ML <- as.matrix(result_ML) %>% dplyr::na_if("")
#result_ML[result_ML == ""] <- NA
#result_ML <- as.matrix(result_ML) %>% dplyr::na_if("")


# Assign the independent and dependent covariates
x_xgb <- data.matrix(dat_XGB[, c(full_covariate_xgm)])
y_xgb <- log(dat_XGB[, i])
plots_list <- list() # Initialize the list to store plots

dat <- dat %>%
mutate(across(all_of(cov_factors), as.numeric))
dplyr::mutate(dplyr::across(dplyr::all_of(cov_factors), as.numeric))
# First case: covariates are selected after the vote
if (is.na(result_ML[i, 1]) == F) {
list_cov <- strsplit(gsub(" ", "", result_ML[i, 1]), ",")
x.selected_final <- as.matrix(dat_XGB %>% select(all_of(list_cov[[1]])))
x.selected_final <- as.matrix(dat_XGB %>% dplyr::select(dplyr::all_of(list_cov[[1]])))

train.ind <- createDataPartition(seq(1, nrow(x.selected_final)), times = 1, p = 0.8, list = FALSE)
train.ind <- caret::createDataPartition(seq(1, nrow(x.selected_final)), times = 1, p = 0.8, list = FALSE)
training <- as.matrix(x.selected_final[train.ind, ])
colnames(training) <- colnames(x.selected_final)
testing <- as.matrix(x.selected_final[-train.ind, ])
Expand All @@ -371,7 +351,7 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac
y.xgm_test <- y[-train.ind, ]

if (length(list_cov[[1]]) != 0 ) {
xgb.mod <- xgboost(
xgb.mod <- xgboost::xgboost(
data = training,
label = y.xgm_train,
nrounds = 200,
Expand All @@ -388,16 +368,16 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac
if (!(any(grepl(k, list_cov[[1]])))) {
if (k %in% cov_continuous) {
data_plot <- data.frame(Residuals = residuals, cov = c(dat[-train.ind, k]))
plot <- ggplot(data_plot, aes(x = cov, y = Residuals)) +
geom_point() +
labs(x = k, y = paste("Residuals", i)) +
geom_smooth(method = 'lm')
plot <- ggplot2::ggplot(data_plot, ggplot2::aes(x = cov, y = Residuals)) +
ggplot2::geom_point() +
ggplot2::labs(x = k, y = paste("Residuals", i)) +
ggplot2::geom_smooth(method = 'lm')
} else {
data_plot <- data.frame(Residuals = residuals, cov = c(dat[-train.ind, k]))
plot <- ggplot(data_plot, aes(x = as.factor(cov), y = Residuals)) +
geom_boxplot() +
geom_point(position = position_jitter(width = 0, height = 0)) +
labs(x = k, y = paste("Residuals", i))
plot <- ggplot2::ggplot(data_plot, ggplot2::aes(x = as.factor(cov), y = Residuals)) +
ggplot2::geom_boxplot() +
ggplot2::geom_point(position = ggplot2::position_jitter(width = 0, height = 0)) +
ggplot2::labs(x = k, y = paste("Residuals", i))
}

# Calculate correlation and p-value
Expand All @@ -409,14 +389,14 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac

# Try different train indices until a valid p-value is obtained or reach the maximum attempts
while (is.na(p_value) && attempts <= max_attempts) {
train.ind <- createDataPartition(seq(1, nrow(x.selected_final)), times = 1, p = 0.8, list = FALSE)
train.ind <- caret::createDataPartition(seq(1, nrow(x.selected_final)), times = 1, p = 0.8, list = FALSE)
training <- x.selected_final[train.ind, ]
testing <- x.selected_final[-train.ind, ]

y.xgm_train <- y[train.ind, ]
y.xgm_test <- y[-train.ind, ]

xgb.mod <- xgboost(
xgb.mod <- xgboost::xgboost(
data = training,
label = y.xgm_train,
nrounds = 200,
Expand Down Expand Up @@ -461,9 +441,9 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac
comptage <- as.data.frame(table(list_cov_nb))
if (nrow(comptage) != 0) {

x.selected_final <- as.matrix(dat_XGB %>% select(all_of(comptage$list_cov_nb)))
x.selected_final <- as.matrix(dat_XGB %>% dplyr::select(dplyr::all_of(comptage$list_cov_nb)))

train.ind <- createDataPartition(seq(1, nrow(x.selected_final)), times = 1, p = 0.8, list = FALSE)
train.ind <- caret::createDataPartition(seq(1, nrow(x.selected_final)), times = 1, p = 0.8, list = FALSE)
training <- as.matrix(x.selected_final[train.ind, ])
colnames(training) <- colnames(x.selected_final)
testing <- as.matrix(x.selected_final[-train.ind, ])
Expand All @@ -475,7 +455,7 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac
y.xgm_test <- y[-train.ind, ]

if (length(list_cov_nb) != 0 ) {
xgb.mod <- xgboost(
xgb.mod <- xgboost::xgboost(
data = training,
label = y.xgm_train,
nrounds = 200,
Expand All @@ -491,16 +471,16 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac
for (k in full_covariate) {
if (k %in% cov_continuous) {
data_plot <- data.frame(Residuals = residuals, cov = c(dat[-train.ind, k]))
plot <- ggplot(data_plot, aes(x = cov, y = Residuals)) +
geom_point() +
labs(x = k, y = paste("Residuals", i)) +
geom_smooth(method = 'lm')
plot <- ggplot2::ggplot(data_plot, ggplot2::aes(x = cov, y = Residuals)) +
ggplot2::geom_point() +
ggplot2::labs(x = k, y = paste("Residuals", i)) +
ggplot2::geom_smooth(method = 'lm')
} else {
data_plot <- data.frame(Residuals = residuals, cov = c(dat[-train.ind, k]))
plot <- ggplot(data_plot, aes(x = as.factor(cov), y = Residuals)) +
geom_boxplot() +
geom_point(position = position_jitter(width = 0, height = 0)) +
labs(x = k, y = paste("Residuals", i))
plot <- ggplot2::ggplot(data_plot, ggplot2::aes(x = as.factor(cov), y = Residuals)) +
ggplot2::geom_boxplot() +
ggplot2::geom_point(position = ggplot2::position_jitter(width = 0, height = 0)) +
ggplot2::labs(x = k, y = paste("Residuals", i))
}

# Calculate correlation and p-value
Expand All @@ -512,7 +492,7 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac

# Try different train indices until a valid p-value is obtained or reach the maximum attempts
while (is.na(p_value) && attempts <= max_attempts) {
train.ind <- createDataPartition(seq(1, nrow(x.selected_final)), times = 1, p = 0.8, list = FALSE)
train.ind <- caret::createDataPartition(seq(1, nrow(x.selected_final)), times = 1, p = 0.8, list = FALSE)
training <- as.matrix(x.selected_final[train.ind, ])
colnames(training) <- colnames(x.selected_final)
testing <- as.matrix(x.selected_final[-train.ind, ])
Expand All @@ -521,7 +501,7 @@ generate_residualsplots <- function(tab, list_pop_param, cov_continuous, cov_fac
y.xgm_train <- y[train.ind, ]
y.xgm_test <- y[-train.ind, ]

xgb.mod <- xgboost(
xgb.mod <- xgboost::xgboost(
data = training,
label = y.xgm_train,
nrounds = 200,
Expand Down
Loading