Out of sample prediction with new category

[f2e2f]

@junpenglao

with varying_intercept:

Thanks, I successfully used the hypermean as suggested. @junpenglao’s suggestion is also interesting and may save me a bunch of code. I might try that next. Thanks all!