Add files via upload

Author: naas94

interpret_results/confusion_correlation.R

@@ -0,0 +1,168 @@
+#Set random seed
+set.seed(123)
+
+#Set working directory
+setwd("C:/Users/adamen/OneDrive - Universitetet i Oslo/documents/Doktorgrad/Artikkel 4/R/")
+
+#Function to aggregate confusion matrices
+confusion.matrix <- function(ED_matrix, confusion_list, elements, progress) {
+  
+  #Create matrix for storing confusion matrices
+  confusion_matrix <- matrix()
+  
+  #Create array for storing confusion matrices
+  confusion_array <- array(matrix(0,ncol(ED_matrix),ncol(ED_matrix)), dim = c(ncol(ED_matrix),ncol(ED_matrix),length(elements)))
+  
+  #Aggregate confusion matrices within the hierarchical level
+  for (k in 1:length(elements)) {
+    
+    #Create template with all classes for storing confusion matrix results
+    confusion_template <- matrix(0,ncol(ED_matrix),ncol(ED_matrix))
+    
+    #Store confusion matrix k
+    confusion_matrix <- confusion_list[[elements[k]]]
+    
+    #Store confusion matrix from one model in the template (that has all the ecosystem types)
+    for (i in 1:ncol(confusion_matrix)) {
+      for (j in 1:nrow(confusion_matrix)) {
+        
+        #Add values from confusion matrix k to the template
+        confusion_template[as.numeric(rownames(confusion_matrix)[j]),as.numeric(colnames(confusion_matrix)[i])] <- confusion_template[as.numeric(rownames(confusion_matrix)[j]),as.numeric(colnames(confusion_matrix)[i])] + confusion_matrix[j,i]
+      }
+    }
+    
+    #Add results from confusion matrix k to the array
+    confusion_array[,,k] <- confusion_template
+    
+    #Print progress
+    if (progress == T) {
+      print(k/length(elements))
+    }
+  }
+  
+  #Aggregate the confusion matrices with the mean
+  mean_confusion_matrix <- apply(confusion_array, c(1,2), mean)
+  
+  #Return the aggregated confusion matrix
+  return(mean_confusion_matrix)
+}
+
+#Specify file paths
+file_paths <- list.files("C:/Users/adamen/OneDrive - Universitetet i Oslo/documents/Doktorgrad/Artikkel 4/ED/", pattern = "xlsx$", full.names = TRUE)
+
+#Create list for class codes
+conversion_schemes <- list()
+
+#Import files with class names and store them in list
+for (i in 1:length(file_paths)) {
+  conversion_schemes[[i]] <- colnames(as.data.frame(readxl::read_xlsx(file_paths[i])))
+}
+
+#Specify file paths
+file_paths <- list.files("C:/Users/adamen/OneDrive - Universitetet i Oslo/documents/Doktorgrad/Artikkel 4/ED/", pattern = "xlsx$", full.names = TRUE)
+
+#Create list for class codes
+conversion_list <- list()
+
+#Import files with class names and store them in list
+for (i in 1:length(file_paths)) {
+  conversion_list[[i]] <- as.data.frame(readxl::read_xlsx(file_paths[i]))
+}
+
+#Import confusion matrices for classifiers
+classifier_confusion <- readRDS("results/classifier_confusion.rds")
+
+#Import confusion matrices for interpreters
+interpreter_confusion <- readRDS("results/interpreter_confusion.rds")
+
+#Import model results
+classfier_data <- read.csv("results/results.csv")[,-c(1)]
+
+#Import model results
+interpreter_data <- read.csv("results/interpreter_results.csv")[,-c(1)]
+
+#Create a data frame with all combinations of interpreters and hierarchical levels
+data_frame_classifier <- interpreter_data[,1:2]
+
+#Create list for storing confusion matrices
+aggregated_classifier_list <- list()
+
+for (h in 1:nrow(data_frame_classifier)) {
+  
+  #Specify the correct conversion key (depending on hierarchical level and map scale)
+  if(data_frame_classifier$hierarchicallevel[h] == "gtype1") {
+    conversion_scheme <- conversion_list[[1]]
+    if(data_frame_classifier$interpreter[h] %in% c("A5","F5","D5","C5","B5","E5","X5")) {
+      conversion_scheme <- conversion_list[[2]]
+    }
+  }
+  if(data_frame_classifier$hierarchicallevel[h] == "htype1") {
+    conversion_scheme <- conversion_list[[3]]
+  }
+  if(data_frame_classifier$hierarchicallevel[h] == "htypegr1") {
+    conversion_scheme <- conversion_list[[4]]
+  }
+  
+  #Identify the rows with the same hierarchical levels and map scales
+  interpreter_rows <- which(classfier_data$interpreter == data_frame_classifier$interpreter[h])
+  hierarchicallevel_rows <- which(classfier_data$hierarchicallevel == data_frame_classifier$hierarchicallevel[h])
+  
+  #Specify the row
+  rows <- interpreter_rows[which(interpreter_rows %in% hierarchicallevel_rows)]
+  
+  #Aggregate confusion matrices for classifiers
+  aggregated_classifier_list[[h]] <- confusion.matrix(conversion_scheme, classifier_confusion, rows, progress = F)
+}
+
+#Create data frame with all combinations of interpreters and hierarchical levels
+data_frame_interpreter <- interpreter_data[,1:2]
+
+#Create list for storing confusion matrices
+aggregated_interpreter_list <- list()
+
+for (h in 1:nrow(data_frame_interpreter)) {
+  
+  #Specify the correct conversion key (depending on hierarchical level and map scale)
+  if(data_frame_interpreter$hierarchicallevel[h] == "gtype1") {
+    conversion_scheme <- conversion_list[[1]]
+    if(data_frame_interpreter$interpreter[h] %in% c("A5","F5","D5","C5","B5","E5","X5")) {
+      conversion_scheme <- conversion_list[[2]]
+    }
+  }
+  if(data_frame_interpreter$hierarchicallevel[h] == "htype1") {
+    conversion_scheme <- conversion_list[[3]]
+  }
+  if(data_frame_interpreter$hierarchicallevel[h] == "htypegr1") {
+    conversion_scheme <- conversion_list[[4]]
+  }
+  
+  #Specify row
+  rows <- h
+  
+  #Aggregate confusion matrices for classifiers
+  aggregated_interpreter_list[[h]] <- confusion.matrix(conversion_scheme, interpreter_confusion, rows, progress = F)
+}
+
+#Create column for correlation estimates for each interpreter hierarchical level combination
+data_frame_classifier$confusion_correlation <- NA
+
+#Loop over all lists with aggregated confusion matrices
+for (i in 1:length(aggregated_interpreter_list)) {
+  
+  #Convert the interpreter confusion matrix for to a vector
+  interpreter_confusion_vector <- as.numeric(aggregated_interpreter_list[[i]])
+  
+  #Convert the classifier confusion matrix for to a vector
+  classifier_confusion_vector <- as.numeric(aggregated_classifier_list[[i]])
+  
+  #Identify confusion matrix cells in which there has been assigned ecosystem types 
+  existent_values <- unique(c(which(interpreter_confusion_vector != 0), which(classifier_confusion_vector != 0)))
+  
+  #Store the correlation between classifiers and interpreter in the data frame
+  data_frame_classifier$confusion_correlation[i] <- cor.test(interpreter_confusion_vector[existent_values], classifier_confusion_vector[existent_values], alternative = "two.sided", method = "pearson")$estimate
+}
+
+#Print the results
+mean(data_frame_classifier$confusion_correlation)
+sd(data_frame_classifier$confusion_correlation)
+