Code and tables uploaded

SA_ASC/code/src/sentiment_analysis/modelling.R

+library(e1071)
+library(foreach)
+library(caret)
+library(dplyr)
+library(UBL)
+library(DMwR)
+
+evaluations <- read.csv("E:/mavis/data/tweets.csv")
+summary.factor(evaluations$IBM.Adap)
+summary.factor(evaluations$GC.Adapt)
+summary.factor(evaluations$MC.Adapt)
+summary.factor(evaluations$X.3)
+summary.factor(evaluations$X)
+
+# Data Preparation ----
+
+evaluations <- evaluations %>% filter(!is.na(TWEET.ID))
+
+evaluations$MC.Orig <- ifelse(evaluations$MC.Orig == "NONE", "NEU", as.character(evaluations$MC.Orig))
+evaluations$MC.Orig <- as.numeric(as.factor(evaluations$MC.Orig))
+
+evaluations$X <- factor(evaluations$X, levels = c("NEGATIVO","NO NEGATIVO"))
+evaluations$X.3 <- factor(evaluations$X.3, levels = c("NEGATIVO","NO NEGATIVO"))
+levels(evaluations$X) <- c("NEGATIVO","NO_NEGATIVO")
+levels(evaluations$X.3) <- c("NEGATIVO","NO_NEGATIVO")
+
+col_id_targets <- c(17, 21)
+
+# select data set
+
+#col_id_predictors <- c(8,10,12) # orig
+#col_id_predictors <- c(9, 11, 13) # adapt
+#col_id_predictors <- c(8, 9, 10, 11, 12, 13) # both
+
+#df <- evaluations[,c(col_id_predictors, col_id_targets)]
+
+#df <- df %>% select(-X.3) # 5 EV
+#df <- df %>% select(-X) # 3 EV
+
+#names(df)[ncol(df)] <- "target"
+
+#final_results <- modelate(df, "3 EV", "ORIG", "")
+#final2_results <- data.frame()
+
+conj = c(1:10)
+evs = c('3 EV',"5 EV")
+preds = c("ORIG","ADAP","BOTH")
+
+for(ev in evs){
+for(p in preds){
+  for (c in conj){
+  method = paste("DOWN-rndm", c)
+  final2_results <- rbind(final2_results, modelate(evaluations, ev, p, method))
+}}}
+
+final_results <- rbind(final_results, modelate(evaluations, "5 EV", "ORIG", " DOWN-clust"))
+
+fn <- final_results[!grepl("UP-rndm", final_results$method),]
+final_results <- fn
+
+modelate <- function(df, evaluadores_label, predictores_label, extra_label){
+  
+  if(predictores_label == "ORIG"){print('ORIG');col_id_predictors <- c(8,10,12)}
+  else if(predictores_label == "ADAP"){print('ADAP');col_id_predictors <- c(9, 11, 13)}
+  else if(predictores_label == "BOTH"){print('BOTH');col_id_predictors <- c(8, 9, 10, 11, 12, 13)}
+  
+  df <- df[,c(col_id_predictors, col_id_targets)]
+  
+  if(evaluadores_label == '5 EV'){print('5 EV');df <- df %>% select(-X.3)}
+  else if(evaluadores_label == '3 EV'){print('3 EV');df <- df %>% select(-X)}
+  
+  names(df)[ncol(df)] <- "target"
+  # data prep
+  
+  #set.seed(1)
+  df <- df %>% filter(complete.cases(df))
+  df$id <- seq(1,nrow(df),1)
+  ds <- df
+  
+  ds <- ds %>% select(-id)
+  
+  # tecnica de balanceo
+  set.seed(round(runif(1, 0, 2000)))
+  
+   ds <- under_training <- ds %>% group_by(target) %>% sample_n(128) # 142 numero de negativos por 5 evaluadores
+    
+   # over_training <- ds[ds$target=="NEGATIVO",]
+   # ids <- runif((200/100) * nrow(over_training), 1, nrow(over_training))
+   # over_training_plus <- over_training[ids, ]
+   # ds <- over_training <- rbind(over_training, over_training_plus, ds[ds$target!="NEGATIVO", ])
+  
+   # under_training <- ds[ds$target=="NO_NEGATIVO",]
+   # under_training <- under_training %>% select(-target) %>% filter(complete.cases(under_training))
+   # under_training <- kmeans(under_training, 441)
+   # under_training <- as.data.frame(under_training$centers)
+   # under_training$target <- "NO_NEGATIVO"
+   # ds <- under_training <- rbind(under_training, ds[ds$target!="NO_NEGATIVO", ])
+   
+   # over_training <- SMOTE(target ~ ., data  = ds, perc.over = 200) 
+   # over_training <- over_training[over_training$target=="NEGATIVO",]
+   # ds <- over_training <- rbind(over_training, ds[ds$target!="NEGATIVO", ])
+  
+   # ds <- AdasynClassif(target~., ds, beta=1)
+  
+  inTraining <- createDataPartition(ds$target, p = 0.50, list = TRUE)
+  training <- ds[ inTraining$Resample1,]
+  testing <- ds[ -inTraining$Resample1,]
+  
+  # algorithm application
+  
+  fitControl <- trainControl(
+    method = "repeatedcv",
+    number = 10,
+    repeats = 3,
+    classProbs=TRUE,
+    summaryFunction = twoClassSummary
+  )
+  
+  methods = c('rf', 'C5.0', 'svmLinear' ,'bayesglm', 'LogitBoost', 'mlpWeightDecayML')
+  
+  results <- foreach(method = methods,
+                     .combine = 'rbind') %do% {
+                       print(method)
+                       model <- train(target ~ ., data = training, 
+                                      method = method, 
+                                      trControl = fitControl, 
+                                      preProc = c("center", "scale"),
+                                      metric="ROC",
+                                      tuneLength = 3)
+                       
+                       rocs = model$resample$ROC
+                       
+                       repetition1 = mean(rocs[1:10])
+                       repetition2 = mean(rocs[11:20])
+                       repetition3 = mean(rocs[21:30])
+                       mean_acc = mean(c(repetition1, repetition2, repetition3))
+                       sd_acc = sd(c(repetition1, repetition2, repetition3))
+                       
+                       method_label = paste(method, extra_label)
+                       data.frame(
+                         evaluadores = evaluadores_label,
+                         predictores = predictores_label,
+                         method = method_label,
+                         mean_acc = round(mean_acc,2), 
+                         sd_acc = round(sd_acc,2)
+                         )
+                     }
+  
+  print(results)
+  results
+} 
+
+write.csv2(final2_results, 'E:/mavis/experiment_results_only_RNDM.csv', row.names = F)
+save.image('E:/mavis/experiments.RData')
+load('E:/mavis/experiments.RData')
+
+
+method = 'rf DOWN' 
+ev = '5'
+
+mean(final2_results$mean_acc[grepl(method, final2_results$method)&
+                               grepl("ORIG",final2_results$predictores)&
+                               grepl(ev,final2_results$evaluadores)])
+
+
+mean(final2_results$mean_acc[grepl(method, final2_results$method)&
+                               grepl("ADAP",final2_results$predictores)&
+                               grepl(ev,final2_results$evaluadores)])
+
+
+mean(final2_results$mean_acc[grepl(method, final2_results$method)&
+                               grepl("BOTH",final2_results$predictores)&
+                               grepl(ev,final2_results$evaluadores)])

SA_ASC/code/src/utils/functions.R

+
+install_and_load_dependencies <- function(libs){
+  installed_packages <- installed.packages()
+  
+  installed_packages.names <- installed_packages[,1]
+  
+  for(lib in libs) {
+    if(lib %in% installed_packages.names){ library(lib, character.only = TRUE) }
+    else{
+      install.packages(lib)
+      library(lib, character.only = TRUE) 
+    }
+  }
+}