Life In 19x19

Biggest lesson for me, up to now, is this:

about 70% of the “WRONG!” messages I get are because of me focusing on one corner while the fire’s in another corner

—> I need to look everywhere first before I re/act too quickly, #wholeboard


Then again, this lesson is not new to me. It’s only that I seem to have to repeat classes


Grtz, Tom

Bonobo wrote:Biggest lesson for me, up to now, is this:

about 70% of the “WRONG!” messages I get are because of me focusing on one corner while the fire’s in another corner

—> I need to look everywhere first before I re/act too quickly, #wholeboard

I get this too, but I think part of it may come from not being part of the flow of the game, since nothing is numbered.

For me its very rarely about playing in the wrong area of the board. Well, maybe occasionally if its about "where to spend sente", or which approach to play. But as for defensives moves the order of importance of different moves usually seems rather obvious.

Kanin wrote:I had an idea as I was browsing through the discussion about some 'wrong' moves that might actually be not so bad. You could create a system where 'highly contested' positions will be saved somewhere for your viewing. Like, let's say in a particular position a sufficiently high number of players will play the same 'wrong' move. Then you get a message about it and you can view that position. Then you (or someone else/a few others) could either a) let the program know that this is actually a good move and change it from 'wrong' to 'right' or b) make a note as to why this is not a good move.

Of course, for this you may need some help from stronger players. But maybe you could try it with a decently high threshold for what counts as a 'highly contested' position. If the positions that the system feeds you are not too many and/or too frequent, at least I wouldn't mind having a look at them and passing them on to pro players if I can't decide whether it's a good move (a), or why it's a bad move (b).

I think in the long run this could improve the teaching potential of the program a lot, especially when players can get advice on why their choice was a mistake.

It already does this. Its the reason why it states the strength of those who the correct answers are based on. It could well be for example 7.5dan+ for some problem if there are lots of answers, but very few strong answers, then it cant afford to be picky and accepts slightly weaker answers as "correct" since so many weaker players will benefit from the increased diversity of answers in that problem.

Looks like a very useful tool for grinding opening positions - i like it!

And here's what I think would make it even better:

1) After a right answer, don't jump to next problem right away. Instead, have a big "next" button on the side and show all the options like with a wrong answer. Only move on after I decide to click on the button, or perhaps just somewhere on the board. (I think this is pretty big, the following are not as important)

2) After I signed up, the text on the right side doesn't reset between problems and keeps scrolling (did it do that in the first place?) and is maybe a bit too much of a wall of text. Have it clear the area after each problem, so that it shows the standard "your ranke is" etc. only for the current problem.
3) On my computer (newest Firefox) the board position is very slightly off. There's a small part of black visible on the bottom (this looks normal), but on the top my bookmark toolbar covers a few pixels of the board's edge. I can also keep scrolling up (because of the wall of text on the right) and as the board is at the bottom of the page, it will disappear.
4) Not sure if feasible to implement, but browsing different paths that occurred in pro games directly on your website would be really cool. (i.e. after I answer, I can keep clicking on the next move that took place in pro game X)
5) The link to pro games, while it opens up in go4go.net, doesn't seem to actually have the position I just saw, and it opens up somewhere in end game each time.

In any case, great work and regardless of the above, I'll use it for some fuseki study from now on.

walpurgis wrote:Looks like a very useful tool for grinding opening positions - i like it!

+1

And here's what I think would make it even better:

1) After a right answer, don't jump to next problem right away. Instead, have a big "next" button on the side and show all the options like with a wrong answer. Only move on after I decide to click on the button, or perhaps just somewhere on the board. (I think this is pretty big, the following are not as important)

+1

2) After I signed up, the text on the right side doesn't reset between problems and keeps scrolling (did it do that in the first place?) and is maybe a bit too much of a wall of text. Have it clear the area after each problem, so that it shows the standard "your ranke is" etc. only for the current problem.

I also noticed this change … did it happen today? And I ditto what walpurgis says about the wall of text.

3) On my computer (newest Firefox) the board position is very slightly off. There's a small part of black visible on the bottom (this looks normal), but on the top my bookmark toolbar covers a few pixels of the board's edge. I can also keep scrolling up (because of the wall of text on the right) and as the board is at the bottom of the page, it will disappear.

confirmed with Chrome/OS X

4) Not sure if feasible to implement, but browsing different paths that occurred in pro games directly on your website would be really cool. (i.e. after I answer, I can keep clicking on the next move that took place in pro game X)

+1

[..] In any case, great work and regardless of the above, I'll use it for some fuseki study from now on.

Same here. I only hope I don’t learn wrong moves for my strength … not sure about how to answer follow-up moves.


Greetings, Tom

walpurgis wrote:1) After a right answer, don't jump to next problem right away. Instead, have a big "next" button on the side and show all the options like with a wrong answer. Only move on after I decide to click on the button, or perhaps just somewhere on the board. (I think this is pretty big, the following are not as important)

Not sure about this. I rather ensure a fast experience to those who want it. Especially now that there is the "possible answers" link where you can go look at the answers if it really interests you for that position.

walpurgis wrote:2) After I signed up, the text on the right side doesn't reset between problems and keeps scrolling (did it do that in the first place?) and is maybe a bit too much of a wall of text. Have it clear the area after each problem, so that it shows the standard "your ranke is" etc. only for the current problem.

Working on this...

walpurgis wrote:3) On my computer (newest Firefox) the board position is very slightly off. There's a small part of black visible on the bottom (this looks normal), but on the top my bookmark toolbar covers a few pixels of the board's edge. I can also keep scrolling up (because of the wall of text on the right) and as the board is at the bottom of the page, it will disappear.

Fixed.

walpurgis wrote:4) Not sure if feasible to implement, but browsing different paths that occurred in pro games directly on your website would be really cool. (i.e. after I answer, I can keep clicking on the next move that took place in pro game X)

Unlikely to happen, sry.

walpurgis wrote:5) The link to pro games, while it opens up in go4go.net, doesn't seem to actually have the position I just saw, and it opens up somewhere in end game each time.

Fixed.

It now contains 15000 pro games, all from past 5 years.

Ootakamoku wrote:

walpurgis wrote:2) After I signed up, the text on the right side doesn't reset between problems and keeps scrolling (did it do that in the first place?) and is maybe a bit too much of a wall of text. Have it clear the area after each problem, so that it shows the standard "your ranke is" etc. only for the current problem.

Working on this...

I have something working now to address this, let me know what you think.

This is comfortably the most addictive go site ever.

quantumf wrote:This is comfortably the most addictive go site ever.

Yes, I also find I have to drag myself away from it (proof: mouse claw scratches on my desk) in order to get anything done (okay, much of “anything” is procrastination anyway, but … ).

And meanwhile it estimates my rank to be “between 15.4 Kyu and 13.5 Kyu” <sob>

Greetings, Tom

Ootakamoku wrote:Increasing the rank for it mitigates the problem, hopefully to the point that it becomes irrelevant on the larger scale. In any case, ideal situation would be to know exact point loss from each move, compared to the best possible move in the given situation. And use that to judge the players fuseki skill. However having no such data available, I make do with what is almost binary data, but it seems to still yield accurate result with increased sample size. Just like in go, we can play better or worse and anywhere in between, yet the end result is almost binary, you either win or you lose. Here we are working backwards, we know the move was almost perfect, or it was not perfect, yet this distinction multiplied by many others from different situations adds up to quite accurate prediction of eventual skill.

This is fine if you're evaluating a whole game, but when you are evaluating a position as open as a fuseki, you can't calculate all possible variations to ensure that a move is a winning move, keeps the game in contention, or a losing one easily.

I think it would make more sense to classify a different problem's various answers by rank. A 10k needs to play in the right corner, a 5k needs to discriminate between pincer and extension, and as skill goes up, finer distinctions are made.

This doesn't seem to address the 'fuseki' problems that primarily seem to address how to continue a particular joseki, though.

Wonderful tool, I am impressed!

I do notice that sometimes when my choices are considered wrong, it looks a lot like that move is just not current fashion.

Can you say what the minimum sample size is? How often does a position have to appear in a game before you include it in your test set?

It might be interesting to extend your game record set to include strong amateur records, e.g those from KGS: http://www.u-go.net/gamerecords/

You could then mark moves played by strong amateurs (but not by professionals) as "not wrong". If a player can find KGS 6d+ level moves, I would consider that quite good, even if they do not occur in professional play.

HermanHiddema wrote:I do notice that sometimes when my choices are considered wrong, it looks a lot like that move is just not current fashion.

I have had a similar experience.

HermanHiddema wrote:You could then mark moves played by strong amateurs (but not by professionals) as "not wrong". If a player can find KGS 6d+ level moves, I would consider that quite good, even if they do not occur in professional play.

What about making this an option that the user selects? For example, if I want to practice "modern pro fuseki", then I can set that in my preferences. Then maybe the game collection only includes recent pro games. Or if I want to include "all pro games", "6d+ games", etc., the sample set would be selected accordingly.

This might be nice, because to a 5k, it would be great to play at amateur 6d level, but if a 7d player is playing, he might only be interested in pro moves.

Ootakamoku wrote:

walpurgis wrote:1) After a right answer, don't jump to next problem right away. Instead, have a big "next" button on the side and show all the options like with a wrong answer. Only move on after I decide to click on the button, or perhaps just somewhere on the board. (I think this is pretty big, the following are not as important)

Not sure about this. I rather ensure a fast experience to those who want it. Especially now that there is the "possible answers" link where you can go look at the answers if it really interests you for that position.

well, at least add a small delay so the user can see that he actually got a correct answer. It switches so quickly, it leaves the user wondering what just happened.