« first day (2650 days earlier)      last day (2375 days later) » 

5:16 PM
heh, @gung nowadays you might just be better off with a multi-output NN for multi-class prediction
it's becoming really apparent to me that without a "cost of misclassification" framework available, probability models are really weak for predicting classes
 
 
2 hours later…
7:27 PM
@Glen_b (or anyone else): Any thoughts on stats.stackexchange.com/q/314425/17230? Is it as unclear as I think? Or can you help clarify it?
 

« first day (2650 days earlier)      last day (2375 days later) »