Dear all,
I have a set of data in this form:
> str <data>
'data.frame': 1574 obs. of 14 variables:
$ serial: int 12751 14157 7226 15663 11088 10464 1003 10427 11934 3999 ...
$ plate : int 43 46 22 50 38 37 3 37 41 11 ...
$ well : int 79 333 314 303 336 96 235 59 30 159 ...
$ sample: int 266 295 151 327 231 218 21 218 249 84 ...
$ target: chr "HEV 2-AI5IQWR" "Dientamoeba fragilis-AIHSPMK" "Astro
2 Liu-AI20UKB" "C difficile GDH-AIS086J" ...
$ ori.ct: num 0 33.5 0 0 0 ...
$ ct.out: int 0 1 0 0 0 0 0 1 0 0 ...
$ mr : num -0.002 0.109 0.002 0 0.001 0.006 0.015 0.119 0.003 0.004 ...
$ fcn : num 44.54 36.74 6.78 43.09 44.87 ...
$ mr.out: int 0 1 0 0 0 0 0 1 0 0 ...
$ oper.a: int 0 1 0 0 0 0 0 1 0 0 ...
$ oper.b: int 0 1 0 0 0 0 0 1 0 0 ...
$ oper.c: int 0 1 0 0 0 0 0 1 0 0 ...
$ cons : int 0 1 0 0 0 0 0 1 0 0 ...
from which I have selected two numerical variables correspondig to x
and y in a Cartesian plane and one outcome variable (z):
> df = subset(t.data, select = c(mr, fcn, cons))
> df$cons = factor(c("negative", "positive"))
> head(df)
mr fcn cons
1 -0.002 44.54 negative
2 0.109 36.74 positive
3 0.002 6.78 negative
4 0.000 43.09 positive
5 0.001 44.87 negative
6 0.006 2.82 positive
I created an SVM the method with the KERNLAB package with:
> mod = ksvm(cons ~ mr+fcn, # i prefer it to the more canonical "." but the
> outcome is the same
data = df,
type = "C-bsvc",
kernel = "rbfdot",
kpar = "automatic",
C = 10,
prob.model = TRUE)
> mod
Support Vector Machine object of class "ksvm"
SV type: C-bsvc (classification)
parameter : cost C = 10
Gaussian Radial Basis kernel function.
Hyperparameter : sigma = 42.0923201429106
Number of Support Vectors : 1439
Objective Function Value : -12873.45
Training error : 0.39263
Probability model included.
First of all, I am not sure if the model worked because 1439 support
vectors out of 1574 data points means that over 90% of the data is
required to fix the hyperplane. this does not look like a model but a
patch. Secondly, the prediction is rubbish -- but this is another
story -- and when I try to create a confusion table of the processed
data I get:
> pred = predict(mod, df, type = "probabilities")
> acc = table(pred, df$cons)
Error in table(pred, df$cons) : all arguments must have the same length
which again is weird since mod, df and df$cons are made from the same dataframe.
Coming to the actual error, I tried to plot the model with:
> plot(mod, data = df)
> kernlab::plot(mod, data = df)
but I get this error:
Error in .local(x, ...) :
Only plots of classification ksvm objects supported
Would you know what I am missing?
Thank you
--
Best regards,
Luigi
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
