kykim0

Thanks for the feedback! I actually discovered sth quite interesting while trying to compare the run time for the single thread case, which is that when using a single thread,...

It's also worth mentioning that what is implemented in this PR is in one sense node-level locking e.g., backpropagating is done independently for different nodes, but in another sense is...

I decided to bite the bullet and try the alternative design of representing each node in the tree as a separate object. This made the code (much) simpler especially the...

Thanks @WhiffleFish for review! I do remember reading about `SpinLock`, but the documentation seems to suggest using `ReentrantLock` in general. Do you have a better sense of the difference between...

@WhiffleFish Thanks for the explanation! The difference makes a good sense. I'm wondering though whether we can generally assume that a given MDP is acyclic. In the paper you linked,...

Thanks @zsunberg for the review! I agree with your concern, and it does make sense to be more careful with a stable package that people already use. I don't think...

I also realize suggestion #2 and #3 aren't necessarily mutually exclusive. If you think it's worth having a version of vanilla MCTS that's more optimized for multithreading, we can also...

Thanks for the feedback (also for the nice comparison test)! I'll go ahead and start a new package JuliaPOMDP/MonteCarloTreeSearch.jl, and also see if we can do sth quick for #2....

Looks like I do have permission to create a package under JuliaPOMDP but wasn't quite sure what the canonical settings are for JuliaPOMDP packages. Could you share with me sample...

I tried the fix by @WhiffleFish, which was to not unnecessarily grab the node lock in the `best_sanode_UCB` method and also saw a bit of speed up though not as...