I had to add a new feature to the core model called re_sum_zero_std.
This is a standard deviation for the prior that the sum of the random effects for each fixed effect is zero. If you do not specify this option for a variable, infinity is used for the standard deviation. This makes the code backward compatible.
You will also note, if you inspect the example, that i had to change the link function for the alpha variable from exp to log (otherwise the scaling is very poor and the optimizer cannot make progress).
I had to add a new feature to the core model called re_sum_zero_std. This is a standard deviation for the prior that the sum of the random effects for each fixed effect is zero. If you do not specify this option for a variable, infinity is used for the standard deviation. This makes the code backward compatible.
You will also note, if you inspect the example, that i had to change the link function for the alpha variable from exp to log (otherwise the scaling is very poor and the optimizer cannot make progress).