Hey, Alex,
once again thank you for your prompt reply and excuse my late answer 
I tried to send the selection as “int” into the syntax, but it didn’t work either.
If you ask me, I’d guess you’re probably right about the format of my answers (–> the C1_x).
The problem is that there are “zeros” after the transformation. If I understood you correctly, categorical and multinomal sampling methods do not work if there are zeros in the observed dataset. So this won’t be an option.
I’m currently trying to rewrite the code to a more classic version of the conjoint estimate (each alternative is a line where the answer is encoded in 0/1; sampling with Bernoulli), but the tensors give me a headache ;-).
I´ll keep you updated (… and would ask you again if I am not able to solve the tensor issue 
)
Daniel