"[DynamicPPL] attempt to link a linked vi" warning when aborting sampling and returning minus infinity

[jlperla]

I have some code that may have failures and need to reject the sample, which I do with something like

    if !(sol.retcode == :Success)
        @addlogprob! -Inf
        return
    end

This used to work fine (though it gives a warning which doesn't appear to be possible to turn off the last time I looked). In recent DynamicPPL, though, it has additional warnings for each parameter

┌ Warning: The current proposal will be rejected due to numerical error(s).
│   isfinite.((θ, r, ℓπ, ℓκ)) = (true, true, false, true)
└ @ AdvancedHMC C:\Users\jesse\.julia\packages\AdvancedHMC\51xgc\src\hamiltonian.jl:47
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821
┌ Warning: [DynamicPPL] attempt to link a linked vi
└ @ DynamicPPL C:\Users\jesse\.julia\packages\DynamicPPL\1qg3U\src\varinfo.jl:821

Is this an innocuous warning? If so, any way to turn it off (and also the one in AdvancedHMC if that has been resolved.

[torfjelde]

I'm having trouble identifying what change in DPPL is causing this tbh; I'll have to do some digging. I also can't reproduce this issue with an example model, e.g. DynamicPPL.TestUtils.demo_dot_assume_dot_observe() :confused:

It shouldn't cause any issues wrt. correctness though because it just goes "oh, it's already linked; I won't link it then". But stilll weird that it shows up, so I'm slightly worried.

[juliohm]

I am getting the same warning on a model I just started working on. Can you please elaborate on what exactly it means for end users? The message isn't very helpful.

[CiaranOMara]

I'm also getting this warning.

[yebai]

@torfjelde will #417 fix this?

[torfjelde]

No, not that I'm aware of. I think this is an issue in Turing, but needs a bit more digging to find the source.

[yebai]

This is likely resolved. Please re-open if similar issues are observed again.