Fixed bug in Ultimate Tic Tac Toe termination conditions

[armatthews]

Fixed bug in Ultimate Tic Tac Toe implementation that caused games to end when any small board was terminal instead of when all small boards are terminal

[lanctot]

Thanks!

[lanctot]

Regarding the failing tests: seems like a new issue in Dark Hex is causing a compilation problem that wasn't happening before.

I am on vacation at the moment but will take a look when I get back next week. Once we fix it you can just pull the changes from master and retrigger the tests.

[lanctot]

Regarding the failing tests: seems like a new issue in Dark Hex is causing a compilation problem that wasn't happening before.

I am on vacation at the moment but will take a look when I get back next week. Once we fix it you can just pull the changes from master and retrigger the tests.

Ok, this fix has been merged into master. Can you pull changes from master and push the merge commit to trigger the tests again?

[lanctot]

@armatthews just checking to see if you can do this? Would love to give you proper credit for this fix, but I can't import it until the tests are passing.

[PatrickChrestin]

@lanctot
I have an explanation for this error: "Spiel Fatal Error: /Users/runner/work/open_spiel/open_spiel/open_spiel/tests/basic_tests.cc:494 game_length <= game.MaxGameLength()
game_length = 85, game.MaxGameLength() = 81"

It is happening because in the basic_tests.cc the game_length is increased by one when an action is performed.
An action in the game ultimate_tic_tac_toe can be either to make a move in a subboard, when only one specific sub board is available, or selecting a sub board, in which the next move will be made, if the user has the choice.
Meaning, that if the use is given the choice of the board, the game_length is increased twice per actual move in the test.

I was able to analyze the error, however I am not fit in c++, so I do not feel comfortable to attempt any fix for this.

[lanctot]

Good find!

Can you just change this line:

open_spiel/open_spiel/games/ultimate_tic_tac_toe/ultimate_tic_tac_toe.h

Line 96 in 0af9920

return tic_tac_toe::kNumCells * kNumSubgames;

to add a * 2 at the end (before the semicolon)?

[lanctot]

Oh @PatrickChrestin I just noticed you are not the original contributor of this fix.

@armatthews just quickly checking if you can finish this -- otherwise, I can make the change on our side and submit independently.

[PatrickChrestin]

yeah, sorry, I was interested in the fix for the problem.
Changing the possible game length is a fix for the problem, but doesn't it compromise the intention of the attribute?
Instead, wouldn't it be better to either:

1. not increment the counter of game length in the test, when selecting a sub board to play in, or
2. combine the selection of the sub board and the actual field to be performed in a single action?

the MR #1314 adapts the max game length of uttt to the max game length of a sub board times the number of sub boards.
In my opinion, selecting a sub board (without performing a move) should not be counted when considering the game length.