In MapSentinel/sraster:

knitr::opts_chunk$set(echo = TRUE)
library(knitr)
library(kableExtra)

Introduction

I want to select 1000 samples per class. To do so also I will consider to keep the same amount of samples per polygons, or at least the same proportion.

Importing data

stat_metrics = NULL
overall_accuracy = NULL
out_confusion_matrix =NULL

for(iter in 1:30){

file_directory = 'C:\\IPSTERS\\sraster\\ins5\\sampling_6\\majority_rule_classification_pca'
list_files = lapply(list.files(file_directory),function(x){paste0(file_directory,"\\",x)})

group_by = 'Object'
nsamples = 7000


data_random_split = lapply(list_files, sampling_strata, nsamples, group_by)

#===============================
#classification random forest
#===============================
#Since we have 22 polygons per class, I will select 15 for training an 7 for testing. 

data_random_split2 = lapply(data_random_split, function_train_selection)
data_random = do.call("rbind",data_random_split2)

#===============================
#Removing nans
#===============================

data_random2 = remove_na_df(data_random)

train = data_random2[data_random2$type == "Training",-c(1,2,3,4,189)]
test = data_random2[data_random2$type == "Test",-c(1,2,3,4,189)]

#===============================
#Modelling
#===============================
#set.seed(222)
model_rf = randomForest::randomForest(Label~. , data = train,ntree = 500)
print(model_rf)
randomForest::importance(model_rf)

#===============================
#Prediction
#===============================

library(caret)
pred_test = predict(model_rf, test)
result_test = caret::confusionMatrix(pred_test, test$Label)

#===============================
#CONFUSION MATRIX
#===============================

table_confusion = as.data.frame(result_test$table)
table_confusion$iteration = iter
out_confusion_matrix =  rbind(out_confusion_matrix, table_confusion)

#===============================
#Accuracies
#===============================


stat_metrics = rbind(stat_metrics, result_test$byClass[,c(1,3,7)])
overall_accuracy = rbind(overall_accuracy,result_test$overall)

cat("ready",iter)
}

#write.csv(out_confusion_matrix,"confusion_matrix_composites.csv")

#gc()
#rm(list=ls())

#write.csv(stat_metrics,"stat_metrics_150_samples_7000p_composites.csv")
#write.csv(overall_accuracy ,"overall_accuraccy_150_samples_7000p_composites.csv")

Confusion matrix

CF = read.csv('C:\\IPSTERS\\sraster\\Results\\confusion_matrix_composites.csv')
CF$X = NULL

confusion_matrix_final = CF[CF$iteration == 1,]
confusion_matrix_final$Freq = 0 
for(i in unique(CF$iteration)){
  index_c = which(CF$iteration == i)
  confusion_matrix_final$Freq = confusion_matrix_final$Freq + CF[index_c,c("Freq")]
}
confusion_matrix_final$iteration= NULL

#write.csv(confusion_matrix_final,"confusion_matrix_final_composites.csv")

#confusion_matrix_final$Freq = confusion_matrix_final$Freq/20

#write.csv(confusion_matrix_final,"confusion_matrix_final2.csv")