Sunday, December 3, 2017

My rules for giving technical help

I spend a reasonable amount of time fielding questions from friends and acquaintances on how to solve modeling problems in Stan. This can be hugely valuable and enjoyable for both me and the other person. I can't state this enough. Yet it can also be exasperating, especially if the person asking for help hasn't done the necessary ground-work to understand the complex model they want to fit.

So here are my rules on giving help:

  • No syntax questions. This is mostly because other people have asked syntax questions on Stan Users or will benefit from your question in the future. Also, the Stan manual is incredible. 
  • If you haven't started with a simple model first, I won't help. I need to be able to see a fit from a simple model, and all iterations you've made to build up to the complex model you're having trouble with.
  • Similarly, if you haven't simulated your model with fake data (I've written one really easy way to do this here) and tried to estimate the "known unknowns", I won't help. Normally doing this teaches everything you need for most problems. If you've done this and are still having problems (and you've been successful in doing this with simpler iterations of the model in the previous step), then you should ask me stuff! 
If you have done these three things, you'll probably find me to be incredibly happy to help you fit interesting models.

[Edit] The exception is when you're at the start of a modeling project with a "how would one model this?" sort of question. They are enjoyable. 

JS


No comments:

Post a Comment