A chain to vector of NamedTuples utility?

[SamuelBrand1]

Hi everyone!

Functions like condition and generated_quantities take NamedTuples as arguments (and Dicts). Sometimes this is really convenient, for example, in counter-factual modelling. Or in extending samples one model into a related model, e.g. for Bayesian forecasting.

However, I can't find a utility that maps between a chain object and an iterable object of NamedTuples. If it doesn't exist I think this would be a useful addition!

[SamuelBrand1]

PS in comparison to get_params, which is col wise with respect to the chain object, this would be row-wise. It would also recollection vars into vectors if possible e.g. x[1], x[2] etc would be repacked into (x = some_vector,...)

[cpfiffer]

You might be able to use the Tables.jl interface [here](https://github.com/TuringLang/MCMCChains.jl/blob/master/src%2Ftables.jl), though I'm not sure how current it is. Can you try  ``` using Tables rows(chain) ``` or maybe ``` using Tables rowtable(chain) ``` Edit: easily the worst formatted post I've ever made but my phone is not helping me edit it.

[SamuelBrand1]

I added my pass CDCgov/Rt-without-renewal#235 to my project before I saw this.

[cpfiffer]

Cool, love it. Good to close?

[SamuelBrand1]

Would you want me to PR this utility to MCMCChains?

[cpfiffer]

It depends on whether it's materially different from using the Tables.jl interface. If it is, then great, but if it provides roughly the same effect I'm not sure this is the home for it. Could you check the existing tables interface to see if it makes sense?

[SamuelBrand1]

using Tables
rowtable(chain)

Does indeed give an iterable with NamedTuples for samples. Very nice. Definitely using this pattern in the future.

[cpfiffer]

Thanks so much! Appreciate your time.