Shap plots

NAMESPACE

@@ -3,3 +3,7 @@
 S3method(print,mlcov_data)
 export(MLCovSearch)
 export(generate_residualsplots)
+export(generate_shap_summary_plot)
+import(SHAPforxgboost)
+import(ggplot2)
+import(xgboost)

R/cov_search.R

@@ -199,60 +199,48 @@ MLCovSearch <- function(tab, list_pop_param, cov_continuous, cov_factors, seed =
     }
   }
 
-  # Initialize an empty list to store the SHAP summary plots
-  shap_plots <- list()
-
-  # Create a function to generate SHAP summary plots
-  generate_shap_summary_plot <- function(xgb_model, X_train, param_name) {
-    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 + ggplot2::ggtitle(param_name)
-
-    return(p)
-  }
+  # Initialize an empty list to store the SHAP summary data and seed information
+  shap_data <- list()
+  shap_seed <- list()
 
   # Interpretation of Selected covariates Beeswarm Plots
   for (i in list_pop_param) {
     y_xgb <- log(dat_XGB[, i])
-
-    if (is.na(result_ML[i, 1]) == FALSE){
-    list_cov <- strsplit(gsub(" ", "", result_ML[i, 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::xgboost(
-        data = x.selected_final,
-        label = y_xgb,
-        nrounds = 200,
-        objective = "reg:squarederror",
-        verbose = 0
-      )
-
-      # Generate SHAP summary plot for the current parameter
-      shap_plot <- generate_shap_summary_plot(
-        xgb_model = xgb.mod_final,
-        X_train = x.selected_final,
-        param_name = i
-      )
-      shap_plots[[i]] <- shap_plot
-    }
+
+    if (is.na(result_ML[i, 1]) == FALSE) {
+      list_cov <- strsplit(gsub(" ", "", result_ML[i, 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::xgboost(
+          data = x.selected_final,
+          label = y_xgb,
+          nrounds = 200,
+          objective = "reg:squarederror",
+          verbose = 0
+        )
+
+        # Generate SHAP summary plot for the current parameter
+        shap_values <- SHAPforxgboost::shap.values(xgb_model = xgb.mod_final, X_train = x.selected_final)
+        shap_long <- SHAPforxgboost::shap.prep(xgb_model = xgb.mod_final, X_train = x.selected_final)
+
+        # Store shap data and seed
+        shap_data[[i]] <- list(shap_values = shap_values, shap_long = shap_long)
+        shap_seed[[i]] <- .Random.seed
+
+      }
     }
   }
 
-  combined_plots <- gridExtra::marrangeGrob(grobs = shap_plots,nrow = length(shap_plots),ncol = 1)
-
-
-
-
   # Return the result_ML table and the SHAP plots for each parameter
   return(
     list(
       result_ML = result_ML,
       result_5folds = result_5folds,
-      shap_plots = shap_plots,
       list_pop_param = list_pop_param,
-      dat_XGB = dat_XGB
+      shap_data = shap_data,
+      shap_seed = shap_seed
     ) %>% structure(class = "mlcov_data")
   )
 } 

R/generate_shap_summary_plot.R

@@ -0,0 +1,71 @@
+#' Generate SHAP Summary Plots
+#'
+#' This function generates SHAP summary plots for the XGBoost model.
+#' @inheritParams SHAPforxgboost::shap.plot.summary
+#' @param data A list containing required data frames and results.
+#' @param title A character string to customize the title
+#' @param title.position A numeric value from 0.0-1.0 to adjust the alignment of the title
+#' @param ylab A character string to customize the y-axis
+#' @param xlab A character string to customize the x-axis
+#' @return A list of ggplot objects, each representing a SHAP summary plot for a different parameter.
+#'
+#' @examples
+#' # Assuming 'data' is a list with necessary components
+#' \dontrun{
+#' plots <- generate_shap_summary_plot(data, title = "Custom Title", y.inter = 0.25, ...)
+#' }
+#'
+#' @import xgboost
+#' @import ggplot2
+#' @import SHAPforxgboost
+#'
+#' @export
+#' 
+generate_shap_summary_plot <- function(data,
+                                       x_bound = NULL,
+                                       dilute = FALSE,
+                                       scientific = FALSE,
+                                       my_format = NULL,
+                                       min_color_bound = "#FFCC33",
+                                       max_color_bound = "#6600CC",
+                                       kind = c("sina", "bar"),
+                                       title = NULL, 
+                                       title.position = 0,
+                                       ylab = NULL,
+                                       xlab = NULL)
+{
+  # Initialize an empty list to store the SHAP summary plots
+  shap_plots <- list()
+
+  for (i in data$list_pop_param) {
+    if (!is.null(data$shap_data[[i]])) {
+      # Set to the seed used when making the data
+      set.seed(data$shap_seed[[i]])
+
+      shap_values <- data$shap_data[[i]]$shap_values
+      shap_long <- data$shap_data[[i]]$shap_long
+
+      # Generate summary plot
+      p <-
+        SHAPforxgboost::shap.plot.summary(
+          shap_long,
+          x_bound = x_bound,
+          dilute = dilute,
+          scientific = scientific,
+          my_format = my_format,
+          min_color_bound = min_color_bound,
+          max_color_bound = max_color_bound,
+          kind = kind
+        )
+      p <- p + 
+        ggplot2::ggtitle(ifelse(is.null(title), i, title)) +
+        ggplot2::labs(y = ifelse(is.null(ylab), "SHAP value (impact on model output)", ylab), 
+                      x = ifelse(is.null(xlab), "", xlab)) +
+        ggplot2::theme(plot.title = ggplot2::element_text(hjust = title.position))
+      shap_plots[[i]] <- p
+    }
+
+  }
+
+  return(shap_plots)
+}

README.md

@@ -22,6 +22,21 @@ print(result)
 
 ```
 
+Generate SHAP plots:
+
+```
+generate_shap_summary_plot(
+      result,
+      x_bound = NULL,
+      dilute = FALSE,
+      scientific = FALSE,
+      my_format = NULL,
+      title = NULL,
+      title.position = 0.5,
+      ylab = NULL,
+      xlab = NULL)
+```
+
 Generate residual plots:
 
 Cl