In GegznaV/spMisc: Infix Operators, Convenience and Other Kind of Miscellaneous Functions

knitr::opts_chunk$set(echo = FALSE,
                      fig.align = "center",
                      fig.keep  = "all"
                      )

library(tidyr)
library(dplyr)

library(hyperSpec)
library(NMF)

library(ggthemes)
library(xtable)
library(GGally)
library(corrplot)
library(pander)

source('D:/Data/R/R Learning/plotSpFun.R', encoding = 'UTF-8')

library(rpart)
library(randomForest)
library(caret)
library(rattle)

library(doParallel)
cl <- makeCluster(8)
registerDoParallel(cl)


corrMethod <-  "spearman"

Klasifikacija

Šiame pavizdyje naudojamas atsitiktinių miškų (angl. random forests) klasifikatorius (10 tūkst. sprendimų medžių).

Rezultatų patikrinimas vykdomas kryžminio skaidymo į 5 dalis metodu (angl. 5-fold cross-validation).

# load('2016-01-08.RData')
load('2016-01-08 (nRuns = 40).RData')

# Set random seed. Don't remove this line.
set.seed(123)

# k - number of folds
k = 5

# Initialize the accs vector
accs <- rep(0, k)

# Indices for training set
# indeksai cross-validacijai
# 
# 
AMP_obj <- scoresNMF

# AMP_obj    <- AMP_obj[AMP_obj$gr!= "S",,]
# AMP_obj$BoosGr <- as.factor(AMP_obj$BoosGr)

# levels(AMP_obj$Safranin) <- c("0-1","0-1","2-3","2-3")
# 


        BoosGr <- cut(AMP_obj$Boos,breaks = c(8,14,23))
AMP_obj$BoosGr <- BoosGr

pander(as.data.frame(prop.table(table(BoosGr))))

AMP_obj2  <-  AMP_obj;

AMP_obj2$gr <- factor(AMP_obj2$BoosGr)
indices     <- createFolds_strat(AMP_obj2$..[,c("ID","gr")],
                             k = k,
                             returnTrain = TRUE)


 i = 1
# 
# conf  = vector("list", k)
conf2 = vector("list", k)

Rezultatai kiekvienam išskaydymo variantui

for (i in 1:k) {
  print(i)
  # Exclude them from the train set
  training <- AMP_obj[-indices[[i]],]

  # Include them in the test set
  test <- AMP_obj[indices[[i]],]

#   # A model is learned using each training set
#   tree <- rpart(gr ~ spc, training, method = "class")
#   
#   # Draw the decision tree
#   fancyRpartPlot(tree)

  # A model is learned using each training set
  my_forest <- train(BoosGr ~  spc + Safranin,
                       data = training,
                       method = 'parRF',
                       importance = TRUE,
                       ntree = 1000
                     )

  # Make a prediction on the test set using randomForest
  pred <- predict(my_forest,test)


  varImpPlot(my_forest$finalModel)

  # Assign the confusion matrix to conf
  # conf <- table(test$gr, pred)
   conf <- table(test$BoosGr, pred)

  # pander(conf)

  # print(confusionMatrix(test$gr, pred))
  print(confusionMatrix(test$BoosGr, pred)) 

  # Assign the accuracy of this model to the ith index in accs
  accs[i] <- sum(diag(conf))/sum(conf)
}

Bendro tikslumo apibendrinimas

# Print out the mean of accs
pander(data.frame(Pakartojimas = factor(c(1:k, 'vidutinis')),
                  Tikslumas = c(accs,mean(accs))))

Galutinis atsakymas: tikslumas = r mean(accs)

Pastabos

stopCluster(cl)

GegznaV/spMisc documentation built on April 26, 2020, 5:59 p.m.