a tool to analyze and evaluate the role of "luck"
Moderator: Forum Moderators
-
forceOfHabit
- Posts: 8
- Joined: May 5th, 2014, 7:16 pm
a tool to analyze and evaluate the role of "luck"
(first posted here http://forums.wesnoth.org/viewtopic.php ... ck#p570152)
So I understand that there's a steady stream of grumbling about losses due to bad luck in Wesnoth. Has anybody ever tried to write a tool for analysing how lucky each side actually was during the course of the game? I know there's the simple information about actual vs. expected damage, but that is just a fairly primitive sanity check on the PRNG. As one of the earlier posts in this thread (actually the thread where this was first posted) mentions, sometimes the unlikely set of (hits) misses occurs at a point where it has a real impact on the outcome, sometimes it's just an inconvenience that doesn't really matter.
What I'm wondering is, has anyone written a tool to try to distinguish/measure this kind of difference. I think this would be a fun thing to do, and I'm willing to take a stab it, not least because doing it successfully would mean inventing ways to look at the game that are also relevant to writing/improving an AI. After all, if you can accurately conclude that the (non-)death of that orc in the central village was super unlucky because otherwise your side would have won, then you have identified a critical strategic element that your AI should also be aware of.
It would also be kind of cool because it would provide some automated feedback to people posting their replays and asking for analysis/advice. Imagine you could run this tool and it said, yes you were somewhat lucky on turns 6 and 14 but really unlucky on turn 12.
I posted here because in the hope that some of you who have been around for a while might be able to answer the question (has this been tried? i.e is there some prior work I can piggy back off?), and you might also have suggestions for how such a tool might work, what you would like it to do. I'm under no illusions that this would be quick, easy, or satisfy everyone, but I still think it would be cool.
(At the risk of being excessively long winded, this same problem exists with similar games, e.g. Risk. Winning a battle for a country that makes/breaks a continent is more significant than usual, so it would be interesting to have a tool to identify (even after the fact) such crucial moments and how luck impacted them.)
So I understand that there's a steady stream of grumbling about losses due to bad luck in Wesnoth. Has anybody ever tried to write a tool for analysing how lucky each side actually was during the course of the game? I know there's the simple information about actual vs. expected damage, but that is just a fairly primitive sanity check on the PRNG. As one of the earlier posts in this thread (actually the thread where this was first posted) mentions, sometimes the unlikely set of (hits) misses occurs at a point where it has a real impact on the outcome, sometimes it's just an inconvenience that doesn't really matter.
What I'm wondering is, has anyone written a tool to try to distinguish/measure this kind of difference. I think this would be a fun thing to do, and I'm willing to take a stab it, not least because doing it successfully would mean inventing ways to look at the game that are also relevant to writing/improving an AI. After all, if you can accurately conclude that the (non-)death of that orc in the central village was super unlucky because otherwise your side would have won, then you have identified a critical strategic element that your AI should also be aware of.
It would also be kind of cool because it would provide some automated feedback to people posting their replays and asking for analysis/advice. Imagine you could run this tool and it said, yes you were somewhat lucky on turns 6 and 14 but really unlucky on turn 12.
I posted here because in the hope that some of you who have been around for a while might be able to answer the question (has this been tried? i.e is there some prior work I can piggy back off?), and you might also have suggestions for how such a tool might work, what you would like it to do. I'm under no illusions that this would be quick, easy, or satisfy everyone, but I still think it would be cool.
(At the risk of being excessively long winded, this same problem exists with similar games, e.g. Risk. Winning a battle for a country that makes/breaks a continent is more significant than usual, so it would be interesting to have a tool to identify (even after the fact) such crucial moments and how luck impacted them.)