Sampling from chain gives an error

[Shmuma]

Calling sample from the Chains object returns an error "BoundsError: attempt to access 1000×2×1 Array{Float64,3} at index [[954, 537, 1308, 673, 876, 1081, 1182, 1271, 1141, 1315], 1:2, 1:1]"

How to reproduce:

@model function model_height(height)
    μ ~ Normal(178, 20)
    σ ~ Uniform(0, 50)
    height ~ Normal(μ, σ)
end

chain = sample(model_height(178), NUTS(), 1000)
sample(chain, 10)

I guess this happens due to usage of range(chain), which includes the offset to the adaptation steps. If I create sampler as NUTS(0, 0.65), everything works.

Another solution I found is to sample from the result of resetrange(chain), which is better than avoiding adaptation steps. But still, it would be better to fix the problem.

[rspreafico]

With Turing v0.24.2, this no longer errors out

[devmotion]

Indeed, it seems to be fixed. Possibly by #365.