Random.jl

function RandomMDP(ns::Int64, na::Int64, discount::Float64; rng::AbstractRNG=Random.GLOBAL_RNG)
    # random dynamics
    T = rand(rng, ns, na, ns)
    # normalize
    for i = 1:ns, j = 1:na
        T[:,j,i] /= sum(T[:,j,i])
    end
    # random rewards [-0.5, 0.5]
    R = rand(rng, ns, na) .- 0.5
    return TabularMDP(T, R, discount)
end
RandomMDP() = RandomMDP(100, 5, 0.9)


function RandomPOMDP(ns::Int64, na::Int64, no::Int64, discount::Float64; rng::AbstractRNG=Random.GLOBAL_RNG)
    # random dynamics
    T = rand(rng, ns, na, ns)
    # random observation model
    O = rand(rng, no, na, ns)
    # normalize
    for i = 1:ns, j = 1:na
        T[:,j,i] /= sum(T[:,j,i])
        O[:,j,i] /= sum(O[:,j,i])
    end
    # random rewards [-0.5, 0.5]
    R = rand(rng, ns, na) .- 0.5
    return TabularPOMDP(T, R, O, discount)
end
RandomPOMDP() = RandomPOMDP(100, 5, 20, 0.9)