Turing.jl: Argument to pass warmup sample size.

Created on 11 Jun 2018  路  4Comments  路  Source: TuringLang/Turing.jl

I couldn't find any documentation on passing warmup sample size to the samplers, from https://github.com/TuringLang/Turing.jl/issues/265 I understand you use 1k as the default warmup sample size?
It would be a good addition to have a argument that would allow the user to pass in the required warm up sample size.

doc question

Most helpful comment

Thanks for the clarification.

But I guess this will only work for Hamiltonian subtyped methods. Is this correct?

If so we could maybe find a common interface for this purpose. If not we should maybe think about a more convenient interface or kw argument?

All 4 comments

Sorry about our bad documentation. There are actually ways to pass it. Either

  • Constructing samplers with additional arguments by HMCDA(5000, 2000, ...) (the second integer is considered as warmup size), or
  • Passing sa::Stan.Adapt to sample by sample(...; adapt_conf=sa)

Thanks for the clarification.

But I guess this will only work for Hamiltonian subtyped methods. Is this correct?

If so we could maybe find a common interface for this purpose. If not we should maybe think about a more convenient interface or kw argument?

Thanks for the clarification @xukai92. I think having kwarg for this would be a nice feature to have.

Closed by #793

Was this page helpful?
0 / 5 - 0 ratings

Related issues

trappmartin picture trappmartin  路  3Comments

yebai picture yebai  路  6Comments

mateuszbaran picture mateuszbaran  路  5Comments

skanskan picture skanskan  路  5Comments

trappmartin picture trappmartin  路  6Comments