One versus one approach multiclass in r

Question

Supposing My dataset is iris with 3 classes and I want to implement one versus one SVM approach but when i subset the training set for each classifier by class i and class j, I get an empty subset (follow this line #selecting subset of training set where Species equal to class i and class j)

Species <-iris$Species
class <- unique(Species)
set.seed(123)
s<- sample (150,100)
data_train<- iris[s,]
data_test<- iris[-s,]
train <-data_train
test <-data_test
for(i in 2:length(unique(Species))-1){
  for(j in (i+1):length(unique(Species))){
    print(paste(class[i],class[j],sep=","))


    #selecting subset of training set and testing set where coronaryEvent equal to class i and class j
    train <-subset(train, Species %in% c(class[i],class[j]))
   str(train)


  }}
[1] "setosa,versicolor"
'data.frame':   0 obs. of  5 variables:
 $ Sepal.Length: num 
 $ Sepal.Width : num 
 $ Petal.Length: num 
 $ Petal.Width : num 
 $ Species     : Factor w/ 3 levels "setosa","versicolor",..: 
[1] "setosa,virginica"
'data.frame':   0 obs. of  5 variables:
 $ Sepal.Length: num 
 $ Sepal.Width : num 
 $ Petal.Length: num 
 $ Petal.Width : num 
 $ Species     : Factor w/ 3 levels "setosa","versicolor",..: 
[1] "versicolor,virginica"
'data.frame':   0 obs. of  5 variables:
 $ Sepal.Length: num 
 $ Sepal.Width : num 
 $ Petal.Length: num 
 $ Petal.Width : num 
 $ Species     : Factor w/ 3 levels "setosa","versicolor",..:

it's too long to share i am not allowed to share long code here — Rawia Hemdan, Nov 22 '17 at 11:13
Can you then share a fake dataset? Without it it's difficult to reproduce your problem. — Barbara, Nov 22 '17 at 11:14
'data.frame': 213 obs. of 72 variables: $ Diagnosis : int 2 3 3 3 3 2 3 2 3 1 ... $ Indication : int 2 2 2 3 2 2 2 2 2 2 ...$ CoronaryEvent : Factor w/ 6 levels "CABG","Died",..: 3 3 2 3 2 3 3 3 3 3 ... — Rawia Hemdan, Nov 22 '17 at 11:18
It's still not reproducible. Take a look at this https://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example-aka-mcve-minimal-complete-and-ver — Barbara, Nov 22 '17 at 11:24
head(data) Diagnosis Indication Clopidogrel.Clopidogred.dose Time.of.collection..Hours. 1 2 2 4 1 2 3 2 4 2 3 3 2 6 1 4 3 3 4 1 5 3 2 6 1 6 2 2 4 2 — Rawia Hemdan, Nov 22 '17 at 11:37
i wanna subset data by row CoronaryEvent with value of class i and value of class j — Rawia Hemdan, Nov 22 '17 at 11:38
thanks now it works for(i in 1:length(unique(CoronaryEvent))-1){ for(j in (i+1):length(unique(CoronaryEvent))){ print(paste(i,j,sep=",")) print( class[i]) print( class[j]) }} — Rawia Hemdan, Nov 22 '17 at 12:45
for(i in 1:length(unique(CoronaryEvent))-1){ for(j in (i+1):length(unique(CoronaryEvent))){ print(paste(i,j,sep=",")) print(paste(class[i], class[j],sep=",")) }} — Rawia Hemdan, Nov 22 '17 at 12:50

Barbara · Answer 1 · 2017-11-23T11:13:52.250

1

This should be working:

library(dplyr)
set.seed(123)
s<- sample (150,100)
data_train<- iris[s,]
data_test<- iris[-s,]
train <-data_train
test <-data_test
#train$Species <- as.factor(as.character(train$Species))
class <- unique(iris$Species)


for(i in 2:length(unique(iris$Species))-1){

  for(j in (i+1):length(unique(iris$Species))){
    print(paste(class[i],class[j],sep=","))

    class_i <- factor(as.factor(class[i]))
    class_j <- factor(as.factor(class[j]))


    train2 <- rbind(train[match(as.character(train$Species), class_i, nomatch = FALSE), ], 
                    train[match(as.character(train$Species), class_j, nomatch = FALSE), ])

    train2
    str(train2)


  }}

What I did is to assign the iris$Species value directly to class and changed a bit the subset. Let me know if this works as expected.

edited Nov 23 '17 at 11:13

answered Nov 23 '17 at 10:22

Barbara

1,118
2
11
34

Unfortunately i got incorrect results because > table(data_train$Species) setosa versicolor virginica 36 31 33 – Rawia Hemdan Nov 23 '17 at 10:31
the results [1] "setosa,versicolor" 'data.frame': 100 obs. of 5 variables:...[1] "setosa,virginica" 'data.frame': 100 obs. of 5 variables:[1] "versicolor,virginica"'data.frame': 0 obs. of 5 variables: – Rawia Hemdan Nov 23 '17 at 10:32
By this example it should be "setosa,versicolor" a df of 67 observations "setosa,virginica" a df of 69 observations and "versicolor,virginica" a df of 64 observations – Rawia Hemdan Nov 23 '17 at 10:35
@RawiaHemdan updated my answer, now it returns the expected results – Barbara Nov 23 '17 at 11:14

One versus one approach multiclass in r

1 Answers1