CODEWITHSUNDEEP
rr-caret
5xum
5xum

Reputation: 5535

Caret's SVM behaving strangely

I want to create a toy 2D dataset on which to visualize the performance of the SVM algorithm. I produced a dataset which is perfectly separable:

n <- 200;

d<-data.frame(x=runif(n), y=runif(n))

d$class[-1 + 3 * d$x < d$y] <- "1"
d$class[-1 + 3 * d$x > d$y] <- "2"

However, I don't understand the warnings I get when I run the line:

library(caret)

train(class~., data=d, method='svmLinear', tuneGrid=data.frame(C=1))

The method returns a lot of warnings, all of the type:

  23: In data.row.names(row.names, rowsi, i) :
  some row.names duplicated: 3,4,10,12,14,16,18,22,24,27,30,34,35,38,39,45,47,52,54,56,57,64,67,72,74,76,78,81,83,84,87,88,91,92,94,99,103,108,109,111,113,115,117,118,120,121,123,128,129,131,135,137,138,140,141,143,145,146,148,153,164,171,173,178,181,187,189,191,193,194,198 --> row.names NOT used

Furthermore,

Can anyone explain what in the world is happening?

Upvotes: 0

Views: 74

Answers (2)

phiver
phiver

Reputation: 23598

There is nothing wrong with your model and works fine. You can ignore the warning.

This issue was filed under issue 410 on the github page of caret. This is now resolved and will be fixed when a new version is available on cran (version higher than 6.0-70). Or you can download a new version from github.

Upvotes: 1

Kumar Manglam
Kumar Manglam

Reputation: 2832

Well, I am not sure about what is causing the warnings. But when I used other approach to fit model using train function as below:

train(x=d[, c("x", "y")], y=factor(d$class), method='svmLinear', tuneGrid=data.frame(C=1)

I did not get any warnings. The performance of both the methods, is very much same.

Upvotes: 1

Related Questions