We have a large dataset of over 800 resting state scans that were all preprocessed in the same manner. Each subject was run through FEAT preprocessing including motion correction, full registration, and automatic dimensionality estimation with melodic ICA.
After having hand classified 24 subjects into noise/non-noise components and generating the fix feature folder for each of them, for 9 of the 24 subjects we ran into the two following types of errors (found in the hidden .Rlog text files) when we wanted to execute the training command:
> # select using all
> thr.fsc <- quantile(f.vec, q.thr)
> thr.krs <- quantile(p.vec, q.thr)
> thr.glm <- quantile(p.glm, q.thr)
> thr.svm <- quantile(w.svm, q.thr)
Error in quantile.default(w.svm, q.thr) :
missing values and NaN's not allowed if 'na.rm' is FALSE
Calls: quantile -> quantile.default
Execution halted
> # # (1)
> train.data <- hcp.data
> # SVM
> svm.rbf1 <- svm(class.labs ~ ., data = train.data, probability = T, scale = T, kernel = "radial")
> svm.rbf.prd1 <- attributes(predict(svm.rbf1, train.data, probability=T))$probabilities[,2]
Error in names(ret2) <- rowns :
'names' attribute [427] must be the same length as the vector [425]
Calls: predict -> predict -> predict.svm
Execution halted
For the other 15 subjects, there are no errors and we can generate the training file. However, we cannot classify the components into noise/non-noise for the 9 subjects that show these errors and we believe that this will prevent us from classifying other subjects in the larger 800 subject dataset.
Is there any known cause for these errors and how can we fix them without re-preprocessing the entire dataset? What file is used to create the 'names' attribute? Perhaps we can change something in that file to make the vectors the same size?
Thank you in advance for any advice you may have for us!
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|