MCMCThreads error

[wupeifan]

Currently, I set the Turing sampler to be just sampling one chain and it can work, see https://github.com/HighDimensionalEconLab/HMCExamples.jl/blob/main/src/estimate_rbc_1_joint.jl#L34-L40

chain = sample(turing_model, NUTS(num_adapts, d.target_acceptance_rate; max_depth = d.max_depth),
        d.num_samples;
        init_params=[p_d..., ϵ0],
        progress=true,
        save_state=true,
        callback,
    )

If it is replaced by the MCMCThreads() version, i.e.,

julia>     chain = sample(turing_model, NUTS(num_adapts, d.target_acceptance_rate; max_depth = d.max_depth),                                                                                                                             
               MCMCThreads(),                                                                                                                                                                                                            
               d.num_samples,                                                                                                                                                                                                              
               d.num_chains;                                                                                                                                                                                                       
               init_params=[p_d..., ϵ0],                                                                                                                                                                                                     
               progress=true,                                                                                                                                                                                                                
               save_state=true,                                                                                                                                                                                                              
               callback,                                                                                                                                                                                                                     
           )
ERROR: deepcopy of Modules not supported
Stacktrace:
  [1] error(s::String)
    @ Base .\error.jl:33
  [2] deepcopy_internal(x::Module, stackdict::IdDict{Any, Any})
    @ Base .\deepcopy.jl:33
  [3] deepcopy_internal(x::Any, stackdict::IdDict{Any, Any}) (repeats 3 times)
    @ Base .\deepcopy.jl:76
  [4] deepcopy
    @ .\deepcopy.jl:26 [inlined]
  [5] #28
    @ .\none:0 [inlined]
  [6] iterate
    @ .\generator.jl:47 [inlined]
...

@cpfiffer you can replicate this error by changing the codes in https://github.com/HighDimensionalEconLab/HMCExamples.jl/blob/main/src/estimate_rbc_1_joint.jl#L34-L40, and compile the package again, then run this https://github.com/HighDimensionalEconLab/HMCExamples.jl/blob/main/test/runtests.jl#L6 in parciular.

[cpfiffer]

Huh, can't replicate this on my desktop machine (using code from before df6521b). I'll roll my branch forward and see if I can trigger it with the new code.

[wupeifan]

Are we using the same manifest file? I'm using exactly what is in main branch.
If it only works for you, then could you please post the '] st' results so I check the difference?

[cpfiffer]

Are we using the same manifest file? I'm using exactly what is in main branch.
If it only works for you, then could you please post the '] st' results so I check the difference?

Yep, exactly the same manifest.

[cpfiffer]

Replicated using current main -- curious! Will investigate more.

[cpfiffer]

Found it, opened an issue TuringLang/AbstractMCMC.jl#89.

[wupeifan]

Is there anything we can do to avoid that?

[wupeifan]

A temporary solution in HighDimensionalEconLab/DifferentiableStateSpaceModels.jl#70 resolves this one