Skip to Content

Monte Carlo Simulations for board games

3 replies [Last post]
Joined: 04/25/2012

Hello all!

I'm working on a couple different games right now, and one thing that I'd really like to be able to do is come up with better estimates for expected game length. From what I can understand, Monte Carlo simulations could provide a lot of data on what games should be averaged at. I know, I could just play the game 1000 times and get the same information, but I'd like to go into playtesting with a bit more info.

Has anyone used a particular tool or program to run Monte Carlo simulations on thier games? If so, did you think it was a good/useful process, or did playtesting invalidate the simulation results?

pelle's picture
Joined: 08/11/2008
It works for testing specific

It works for testing specific small parts of the game. Scripting with python is a good way and about as simple as it gets ( With no prior experience expect a weeknight or two to figure it out from the tutorials. Might be able to dig up some example script to post later.

Entire games or large parts will not work if there are real player decisions involved.

Joined: 01/17/2013
how about playtesting?

If you want to simulate your game, then you need to have a fully-functioning software version of your game. It need not be pretty, but it does need to have all rules/pieces. And as pelle says, if player choices matter, then you'd need to simulate that too, which means writing a reasonable AI. If you want to have an e-version of your game, then you should go for it, because you'll be writing that simulator/AI anyways. Or, if you have a relatively simple ruleset, and you're a good programmer, then it might also be worth it.

Otherwise, I suggest playtesting. I don't think you'd need 1000 tries. Rather, over the natural course of running playtesting sessions for your game, you'll get a feel for game length, and be able to come up with good numbers for estimated game length. You can also play the game against yourself. It's less fun than with other people, but it can also be informative.

Joined: 04/25/2012
Monte Carlo is more or less an AI system

The entire purpose of the Monte Carlo simulation would be to determine models of possible results by substituting a range of values—a probability distribution—for any factor that has inherent uncertainty. This would functionally provide the equivalent of an AI framework.

I hadn't thought of building out portions of it in Python, which I am familiar with, and a quick google search revealed a wealth of supportive code snippets and math libraries. Hopefully I will be up to the programming task, but if anyone has played with spreadsheet based tools (even ones you pay for) that was more where I was asking for advice or recommendations.

I always find that I need 20-30 playtests before I start seeing all the angles of what people's choices can do to a game. If there is a choice point where people only take that option 5% of the time, I may need even more games before I can really understand the impact of that choice. I'd much rather run 1000 simulations, get an idea for what the upper bounds are, and then play out specific scenarios that emerge as problematic. I'm hoping I can cut the playtests needed to determine very basic information like game time, and use them more to focus on specific gameplay elements.

Syndicate content

forum | by Dr. Radut