The package softImpute is rather for continuous variables I think. With continuous variables the first simulations we have done show better results for imputePCA.
I use ncp=2 because the function estimencp_MCA returned a number of dimensions that it too large. So I know that with 0 dimension missing values are imputed with the "mean" of the category, so with 2 dimensions I have more information. Perhaps more information can be used, ie more dimensions, but it is better to impute with less dimensions than with too many dimensions (in this latter case, you add noise in your data).
It is possible to impute the data set with the imputeMCA function on the overall dataset (considering all the variables as active) and then to perform the MCA on the completed data set (with the object completeObs) using the supplementary variables.
+Inma Alvarez It is difficult to help you because I have never seen this error. All your variables are categorical and you have missing values? Best FH
+Inma Alvarez I'm sorry Dr. Husson I tried again and now the error is this Error in apply(tabdisj[, (vec[i] + 1):vec[i + 1]], 1, which.max) : dim(X) must have a positive length Thank you very much for your time
Thanks, very informative. This is a great acompanyment to the book
The package softImpute is rather for continuous variables I think. With continuous variables the first simulations we have done show better results for imputePCA.
I use ncp=2 because the function estimencp_MCA returned a number of dimensions that it too large. So I know that with 0 dimension missing values are imputed with the "mean" of the category, so with 2 dimensions I have more information. Perhaps more information can be used, ie more dimensions, but it is better to impute with less dimensions than with too many dimensions (in this latter case, you add noise in your data).
Is it possible to impute values and also us supplementary variables? I can't seem to figure out how to do this....
It is possible to impute the data set with the imputeMCA function on the overall dataset (considering all the variables as active) and then to perform the MCA on the completed data set (with the object completeObs) using the supplementary variables.
That's really useful, thanks. But when I run my data in the step
nb
+Inma Alvarez
It is difficult to help you because I have never seen this error.
All your variables are categorical and you have missing values?
Best
FH
+François Husson
Yes, all the variables are factors or booleans and all of them have missing values. Thank you very much
+Inma Alvarez
I'm sorry Dr. Husson I tried again and now the error is this
Error in apply(tabdisj[, (vec[i] + 1):vec[i + 1]], 1, which.max) :
dim(X) must have a positive length
Thank you very much for your time
@@inmaalvarez5172 Have you fixed the problem/error? I come across with the same error. Anyone who can help me please?