# # Now running the loops with a seed going from 1 to 100
# for (i in seq(1,100)) {
#
# print(i)
#
# # Set the seed
# set.seed(i)
#
# # Select the sample for the train dataset (80%)
# train_sample <- sample(unique(FCII_meta$Individual),
# size = round(length(unique(FCII_meta$Individual))*0.8),
# replace = FALSE)
# # Select sample for the test dataset (remaining 20%)
# test_sample <- unique(FCII_meta$Individual)[!unique(FCII_meta$Individual) %in% train_sample]
#
# # Exctract sample name and save the annotation file
# FCII_list <- list()
#
# for (pannel in unique(FCII_meta$Tube.number)) {
#
# df <- FCII_meta[FCII_meta$Tube.number == pannel,]
#
# df_train <- df[df$Individual %in% train_sample,]
#
# # Check number of aml in train data
# n_aml <- sum(df_train$Condition == "aml")
# n_normal <- sum(df_train$Condition == "normal")
#
# # Balance the train dataset
# diff <- n_normal - n_aml
#
# df_train <- df_train[!df_train$FCS.file %in% sample(df_train[df_train$Condition == "normal",]$FCS.file, size = diff, replace = FALSE), ]
#
# df_test <- df[df$Individual %in% test_sample,]
#
# FCII_list[[paste("pannel_", pannel, sep = "")]][["train"]] <- df_train
#
# FCII_list[[paste("pannel_", pannel, sep = "")]][["test"]] <- df_test
#
# }
#
# for (pannel in names(FCII_list)) {
#
# tmp <- FCII_list[[pannel]]
#
# for (df in names(tmp)) {
#
# write.csv(file = paste("./data_and_envs/FlowCapII/", pannel, "_", df, ".csv", sep = ""), x = tmp[[df]], row.names = FALSE)
#
# }
#
# }
#
# # Prepare CONDOR object on the train dataset
# condor <- prep_fcd(data_path = "./data_and_envs/Pannel_4/",
# max_cell = 3000,
# useCSV = FALSE,
# transformation = "auto_logi",
# remove_param = c("Time", "InFile"),
# anno_table = "./data_and_envs/FlowCapII/pannel_4_train.csv",
# filename_col = "FCS.file",
# seed = 91)
#
# # Re order the variable to have the prediction for AML
# condor$anno$cell_anno$Condition <- factor(condor$anno$cell_anno$Condition,
# levels = c("normal", "aml"), labels = c("1_normal", "2_aml"))
#
#
# # Train classifier on the training data
# condor <- train_classifier_model(fcd = condor,
# input_type = "expr",
# data_slot = "orig",
# sample_names = "expfcs_filename",
# classification_variable = condor$anno$cell_anno$Condition,
# family = "binomial",
# type1 = "response",
# parallelCore = 1,
# reg = FALSE,
# seed = 91)
#
# # Load test data in a condor object
# condor_test <- prep_fcd(data_path = "./data_and_envs/Pannel_4/",
# max_cell = 3000,
# useCSV = FALSE,
# transformation = "auto_logi",
# remove_param = c("Time", "InFile"),
# anno_table = "./data_and_envs/FlowCapII/pannel_4_test.csv",
# filename_col = "FCS.file",
# seed = 91)
#
# condor_test$anno$cell_anno$Condition <- factor(condor_test$anno$cell_anno$Condition,
# levels = c("normal", "aml"), labels = c("1_normal", "2_aml"))
#
# # Predict the label of the test dataset
# condor_test <- predict_classifier(fcd = condor_test,
# input_type = "expr",
# data_slot = "orig",
# sample_names = "expfcs_filename",
# model_object = condor$extras$classifier_model,
# seed = 91)
#
# # Merge result with original annotation
# anno <- read.csv("./data_and_envs/FlowCapII/pannel_4_test.csv")
#
# tmp <- merge(x = condor_test$extras$classifier_prediction$xNew.Pred.sample, y = anno, by.x = "sample", by.y = "FCS.file")
#
# # Include a column with the predicted labels (according to a probability of 50%)
# tmp$Pred <- ifelse(tmp$y.Pred.s0 >= 0.5, "pred_aml", "pred_normal")
#
# tmp <- tmp[, c("sample", "Condition", "Pred", "y.Pred.s0")]
#
# # Store the results in a list
# collector[[paste0("Pred_seed_", i)]] <- tmp
#
# table <- table(tmp$Condition, tmp$Pred)
#
# # Calculate performance parameters
# P <- sum(table[1,1] + table[1,2]) # Total positive samples
# TP <- table[1,1] # True positive prediction
# FP <- table[2,1] # False positive prediction
#
# N <- sum(table[2,1] + table[2,2]) # Total negative samples
# TN <- table[2,2] # True negative prediction
# FN <- table[1,2] # False negative prediction
#
# SENS <- TP/P # Sensitivity
# SPEC <- ifelse(N > 0, TN/N, 0) # Specificity
# ACC <- (TP+TN)/(TP+FP+FN+TN) # Accuracy
#
# # Store the perfromacne data in a table
# table_collector <- rbind(table_collector, c(P, TP, FP, N, TN, FN, SENS, SPEC, ACC))
#
# # Clean
# rm(anno, condor, condor_test, df_test, df_train, FCII_list, tmp, ACC, df, FN, FP, i, N, P, pannel,
# SENS, SPEC, table, test_sample, TN, TP, train_sample)
#
# # Do a garbage collection to avoid overload of systems memory
# gc()
#
# }
