Add support for root parallelization w/ majority voting

snowfrogdev / macao

A general purpose game playing A.I. framework based on the Monte Carlo tree search algorithm.

MIT License

23 stars 4 forks source link

Add support for root parallelization w/ majority voting #6

Open snowfrogdev opened 6 years ago

snowfrogdev commented 6 years ago

[ ] Design a paralellism interface
[ ] Implement paralellism support for Node children_processes
[ ] Implement paralellism support for Browser Web Workers
[ ] Add root parallelization

http://www.sciweavers.org/read/evaluating-root-parallelization-in-go-325169

flesler commented 5 years ago

Maybe this helps. Here's a pretty simple example of a server and browser workers that receive a large list of numbers and return some back. https://github.com/wagenaartje/neataptic/tree/master/src/multithreading/workers

snowfrogdev commented 5 years ago

@flesler I've hit a bit of a wall when it comes to multithreading Macao and it has to do with the way Workers and Node child_processes work. The main problem I'm having is that Macao's algorithm requires the Macao user to supply functions that allow Macao to 'play' the game. For multithreading to work, I have to find a way to pass these functions from the main thread to other threads. The problem is that multithreading in JS does not share memory and is done thru serialized message passing and you can't serialize a function, ergo I can't pass the functions to other threads. I have yet to find a solution. Do you have any ideas?

flesler commented 5 years ago

Right, also the states are potentially large JSON objects, not just arrays of numbers or bytes that you can optimize for. It's a hard one now that I think of, seems like all the simulation code and Macao integration functions need to be in a worker module and Macao calls it. So the user needs to put their code in a separate module explicitly, it's a more complex API and requirements no doubt