Using AbstractEnv from CommonRLInterface with POMDPs

Specific Domains Machine Learning
21

If we do report it to Julia, here is a MWE:

julia> struct X{T, A<:AbstractArray{T}}
           t::T
           a::A
       end
julia> Vector{X{Int, Vector{Float64}}}(undef, 2)
ERROR: TypeError: in X, in A, expected A<:(AbstractArray{T, N} where N), got Type{Vector{Float64}}
Stacktrace:
 [1] top-level scope
   @ REPL[15]:1
22

@zsunberg Your fix worked. Thank you!

I will figure out how to submit a ticket to Julia, thanks for the MWE!

For now, I am going to mark the comment about changing the return type to float32 as the fix since it can be done on the user’s side. That will help people in the short term resolve this. In the long term, your PR to DeepQLearning will prevent this from happening here, and hopefully Julia makes that error message more clear.

23

Here is the issue for Julia: https://github.com/JuliaLang/julia/issues/42401
Feel free to add anything I missed. They will probably value your input because you actually know what you are doing :smiley:
Thanks again! You saved me so much time!