Magarena 1.25

[melvin]

Download Magarena 1.25 for Windows/Linux
Download Magarena 1.25 for Mac

1.25 marks the twelfth release of Magarena on http://magarena.googlecode.com, the first release was 1.14 back in 28th May 2011. Since then we've had one release on the last week of each month.

April is a slow month for Magarena. We fixed two bugs, improved the Linux launch script thanks to a user suggestion, and improved the Minimax AI.

The following lists the changes found in 1.25:
- fixed unreachable buttons on duel panel for smaller screens
- fixed bug with exalted by making the declare attackers step follow the rules more closely
- improved Linux launch script thanks to suggestion from pablo.castellazzi
- improved Minimax AI

[muaddib]

What in minimax AI improved?

[melvin]

The AI build one subtree for each move. Previous versions of the minimax AI would give up on a particular subtree if the depth was too large or the number of leaves in the subtree is too many. There were several cases where this happens and that subtree was assigned a score of 0 (by default) and it was chosen as the best move, leading to bad moves.

[ubeefx]

In the MiniMax AI the level determines the number of main phases that is used to evaluate an option. So level 6 means that up to six main phases are passed until the score is determined. There was a mechanism to prevent the AI from creating a very deep search tree, usually in complex game situations, which would take very long to evaluate. It would then give up with a default score. There was however another system that first determined the optimal number of main phases to use before evaluating all choices, to prevent this from happening. So even when you pick level 6, in some cases the AI would only use 4 main phases because the tree is otherwise too deep. This is partly the reason why higher levels offer lesser improvements in complex sitatuations. The option that was used to search for the optimal depth was deliberately not the pass option but the first other option. Melvin changed this too. It is hard to tell what the effect of these improvements is generally, but it might be good in some cases.

[melvin]

Hmm, now that you mentioned it, I realized that my modifications of the main-phase-determination method may have broken it. Previously, the method is to start with the #main phases = AI level and if it led to the AI giving up, reduce #main phases and check again.

Now the method is to let the allowed time be (AI level in seconds)/(number of moves), let #main phases be 1 and increase by 1 while there is time. This method ignores the exception the AI throws when it gives up!

I think the fix in 1.25 to make the AI never give up fixes the bug introduced in the earlier modification to the main-phase-determination method.

[ubeefx]

Maybe time for some nostalgia to celebrate the twelfth release of the official fork.
This is a screen made with the executable built a couple of weeks before Magarena became open source.
It was still called differently at that time.

[beholder]

Awww, that's pretty cool ubeefx. It even has an undo count!?

So I've been away for a bit and needed a break from programming as well. I'll try to start working on Magarena again soon enough. We can't let this great piece of software die after all.

[Huggybaby]

Yes, IIRC yours truly made the feature request for unlimited undos, first implemented in a special edition, so the count went away. And that was just ONE of my FR's that made it into the finished product.

But only bug reporters get credit, not feature requesters. BOO!

[ubeefx]

Yes, you are right about that. But requesters do get the feature they want.
Originally the number of times you could undo was a setting and part of the "difficulty".
It was actually a feature that it was limited and did not need any work to make it unlimited.