SMC test fails with Turing master

[mohdibntarek]

This test (https://github.com/TuringLang/DynamicPPL.jl/blob/master/test/compiler.jl#L333) fails with Turing master. I am not sure why.

[devmotion]

All CI tests passed, it seems?

[devmotion]

Ah wait, I missed that you talk about Turing master.

[devmotion]

BTW this test should not be part of DynamicPPL but moved to Turing. It tests functionality of the inference algorithms and how the chain is constructed, which is not implemented in DynamicPPL.

[mohdibntarek]

I agree. Let's move it there.

[devmotion]

Maybe the values changed because of TuringLang/Turing.jl#1237 (comment) (now the constant terms related to the number of particles are subtracted).

[mohdibntarek]

I am not too familiar with particle samplers. @yebai any idea?

[devmotion]

I'll have a look and transfer the test.

[devmotion]

I'm pretty sure I figured out the problem (it seems to be about not updating the logarithmic weights correctly when sampling with the ESS threshold which is the default - actually I mentioned this potential problem in TuringLang/Turing.jl#1242 (comment)), I'll run more tests and fix the implementation.