Error in { : task 3 failed - "Stopping"

[Vytas Vytautas]

I tried to use RRegrs with my data set, here my data set file:
http://www.mediafire.com/download/x8zv3ikn0ewujkc/ds.GintrCAI_RDKit.csv
With:
> IntrCA1_RDKit<-RRegrs(DataFileName="ds.GintrCAI_RDKit.csv", PathDataSet="pabandymas1_CAIdG_su_RDKit_only", noCores=4, iSplitTimes=1, noYrand=2)
And I get error:
Error in { : task 3 failed - "Stopping"
Here full error:
http://pastebin.com/CxfHq291

Thanks for help.

[muntisa]

I confirm your error. We need to check the formulas for Rsqured and RMSE in SVM because in the case of datasets with the number of features too large compared with the number of cases, there are strange results. 

If you try the same dataset but with only the first 13 features, no error will occur.

[cafernandezlo]

Dear Vytas,

the size of your dataset is the following: 28 samples and 195 features. Much of your features are noisy or redundant, so a complex model such as SVM-RFE is not able to find 64 good features to build the model as you can see in your error.

1. This package has pre-processing options in order to reduce noise and redundant features by removing correlated features and non zero variance features. Use these options in order to remove as much features as you can. Flags: fRemNear0Var="T",fRemCorr="T"

After this process your dataset will be 28 x 66.

2. Please, fell free to read the following reference regarding C and eps parameters as the error is probably indicating that those values are not good for finding support vectors, you need to search for new ones.

http://www.svms.org/parameters/

Default are: RFE_SVM_C="1;5;10;15;20",RFE_SVM_epsilon="0.01;0.1;0.3"

3. Please, sometimes consider to normalize or standardize your dataset. Flag: fScaling="T"

4. Could you please report caret version? as you can see in my trace I manually follow all the steps aforementioned with no errors using version 6.0-52

 packageVersion("caret")

[1] ‘6.0.52’

http://pastebin.com/PgT5zAGV

[Vytas Vytautas]

Thanks for suggestions, I will try them soon.

Here my caret version:

> packageVersion('caret')
[1] '6.0.64'

[cafernandezlo]

ummm this is probably another bug with this version of the Caret Package.

please, do the following:

remove.packages(c('caret'))

library(devtools)

#install Caret Package version 6.0.52, without the bug
install_version(package='caret',version='6.0-52',repos='http://cran.r-project.org')

re-run your experiment or do as mention here http://pastebin.com/PgT5zAGV in order to check this method alone. And, of course, report your results!!

[Vytas Vytautas]

Yes, it works for me without error now.