Marc Lanctot
Marc Lanctot
@elkhrt did we resolve this?
Hmm, I don't know Quordior. Could the movement IDs correspond to the different locations on the board? What move did you take from the first to the second state? @tewalds
Either way, it _should not_ be the case that the same moves are not represented by the same integers (moves should be have state-independent meanings). Like, in chess, a1a2 is...
I think you may have misunderstood. Let's take a game like Tic-Tac-Toe. There are 9 unique moves, each corresponding to one spot on the 3x3 square. At any state in...
> Because now I can't associate a given board state with a common action. It's much worse if the action ids are very strongly dependent on the board. Like if...
> While I can see the logic of that design choice, it massively blows up the action space, no? As opposed to what, though? I don't see any other viable...
> Since that's the design pattern for all of OS, then this isn't a bug. So I'll close this issue shortly. Please don't yet because it's not resolved. I have...
Ok, perfect, understood.. this clarifies a lot. For the case of quoridor: the movement actions should be separate from the wall actions and there should indeed be only 4 of...
@tewalds based on my reading of the rules on here: https://en.m.wikipedia.org/wiki/Quoridor It seems that all those cases can get covered in just 4 movement actions, right? You only jump when...
Np, this is quite an easy change on our side so I will reopen the issue as a reminder and hopefully we can get to it at some point.