return correct probability with RandomPolicy and MultiThreadEnv (#82)

src/implementations/policies.jl

@@ -40,6 +40,7 @@ RandomPolicy(::FullActionSet, env::AbstractEnv, rng) = RandomPolicy(nothing, rng
 # Ideally we should return a Categorical distribution.
 # But this means we need to introduce an extra dependency of Distributions
 # watch https://github.com/JuliaStats/Distributions.jl/issues/1139
+get_prob(p::RandomPolicy{<:VectSpace}, env::MultiThreadEnv) = [fill(1/length(s), length(s)) for s in p.action_space]
 get_prob(p::RandomPolicy, env::MultiThreadEnv) = [get_prob(p, x) for x in env]
 get_prob(p::RandomPolicy, env) = fill(1 / length(p.action_space), length(p.action_space))
 get_prob(p::RandomPolicy{Nothing}, env) = get_prob(p, env, ChanceStyle(env))