It is also important to know if you are ahead or behind in order to choose beween (big but risky move) or (safe but small move)
But, I didn’t see any resource about it.
So, I made quickly prototype in Excel with using RANDBETWEEN( ; ) and drawing it by stacked bar
no komi
I created parameter (area difference less or equal than … )
And so I started to play. When I guess correctly which area is bigger, I decrease parameter. When I mistake, I increase it. I didn’t count, choose intuitively.
Reached max difficulty (area difference 1) surprisingly fast.
So I locked difficulty at max, but after many tries was unable to beat first record of 6 correct guesses in a row. It was not bigger than 2 correct guesses in a row after that, so it seems it was random, not skill.
Maybe someone will make online version of something like this? It certainly easy.
My excel file is too weird to upload. Maybe I will make for myself java version which able to paint more than 2 areas, but I will be too lazy to make it user friendly.
But I saw you already liked and replied there so I guess you know about it
Are you looking for a more “abstract” one with just colored squares? I’ve played around with that on small grids before, but I would have thought it would be way too hard on 19x19, quite impressed that you reached the max difficulty on your version!
But if there are feature requests for a new similar game I’d be happy to spend an evening on it. The most time-consuming part about that old one was I did the positions completely manually to have them be realistic, if I dropped that requirement it would be easier to make multiple difficulty levels, sizes etc.
Yes, I think it may be more interesting to train general skill, not Go related. Just “which geometrical area is bigger?” Randomly generated. And with changing of difficulty like I explained above.
It would be nice to have more groups.
I often misjudge areas when a player has more corners/side and the other, obviously, has more territory in the middle: I realised that I can’t easily eyeball territory difference in that case.
I recently lost a game with about 40 pts difference, but wasn’t able to tell that until I decided to count (slightly before the end of the game). Even looking at it now, after scoring, I still struggle to believe at first sight that white is so far ahead.
So I think this app would be really useful for me.
I’m dreaming about a wide SGF database of real games with territory estimation through whole games (like katago score difference) and picking positions within different ranges, at different stages of each game.
I also think that it would be useful to have stones and empty intersections, not just coloured squares, in order to train the eye with something like the real thing. I think that too is relevant to our perception.
Nice, but way too hard. It asks what exactly score is - so its impossible to answer intuitively. There are no difficulty levels. And rules are Japanese - so Go mechanics are too much integrated, feeling of area difference itself will not be trained as main objective.
It would be fun if one could just call OGS api on finished games, filter by games that weren’t resigned/timeout etc so they went to scoring then call the thumbnail and show that to the player.
I don’t know how to call a list of finished games though, only the thumbnail part. Also one would probably be assuming that the game was scored correctly by the OGS players if taking the game result as the right answer.
(Maybe you can’t since it’d be a big list. Maybe instead running through game ids, or maybe picking a particular tournament with a lot of games)
I will not deter attempts to offer go practice but to me if i feel brave, counting finished games would be my training. Starting on 9x9 maybe. Don’t really see enough advantage to count bars in a chart. Sorry.
Besides a little tool to count precisely in both rules with details is surely a nice feature although i suspect that our SE do that already.
algorithm is easy:
it starts with 100% random white noise picture
then correct number of random dots of certain color are added, so difference between number of white and black dots is as needed
then all single dots surrounded from all sides by opposite color deleted (they become that opposite color)
(if there are too many white dots, white single dot is deleted next,
if there are too many black dots, black single dot is deleted next)
Then thin lines are deleted block by block
then all too small groups to have 2 eyes deleted
in the end dots are attached to survived groups so score is as intended again
I choose score difference and it randomly with 50% chance becomes: white leads or black leads, I don’t know which
try to guess intuitively without actually counting, then see answer
if FAIL, I increase score difference. if OK , decrease it
no komi:
500 tries, separated by columns of 50 on graph
green line: max difference on each of these 50s
red line: average (lower = better too)
so in last 200 attempts I always guessed correctly with at least score difference 11
progress not really seen
So 19x19 is too hard, I need to add ability to decrease board size. And decrease/increase board size instead of changing score difference.
So I will see on which size I able to guess difference 1 perfectly. Then increase/decrease score difference on size bigger than that. So I will be able to progress.
Then instead of asking which color wins, I will try to learn:
(harder) “Is difference 9 or 7?”
(easier) “Is difference 15 or 1?”
I think depending on your rank being able to tell score difference to within 11 points is pretty good. For most kyu ranks I consider any score difference under 20 “undecided” basically right up to scoring… It’s pretty much not until the dan ranks that counting to single digits becomes important.
it means that graph moves mostly random. It decreases with same probability as increases. 11 as max only means that I certainly have more than 50% chance to guess correctly at 11, it doesn’t mean I have close to 100% chance.
just tried on 11 only: 33 wins and 23 loses
So current system makes little sense, I need to change it: I have to guess 2 times in a row on same difficulty correctly to increase difficulty. And 1 error would be enough to decrease difficulty. So if I answer randomly, graph of score difference will increase indefinitely. Ideally graph should reach 1 only if I have more than 50% winrate vs 3.
(missed stones I have to imagine myself as part of training)
I have 4 possible answers now (I’m sure black/white ahead) and (maybe black/white ahead)
if I choose (sure) correctly , difficulty increases by 1
if I choose (sure) incorrectly, difficulty decreases by 3
if I choose (maybe) correctly , difficulty not changes.
if I choose (maybe) incorrectly, difficulty decreases by 1
goal: reach max difficulty
bigger difficulty may mean: bigger board size / smaller area difference / more groups / more complicated shape of groups
Wow! These are really cool projects. Quick estimating (ie, by intuition) is definitely a handy and convenient skill that I’d like to improve at, because I am lazy. Scaling difficulty and realistic positions will be nice features for this kind of training.
