Life In 19x19

daniel_the_smith wrote:I actually use it personally to do pretty broad surveys; I added three things to my queue, one with three moves (a pincer), and two with two moves (two different approaches to 3-4 stone). But I can see what your problem is; if I added a two-space pincer as well as the one-space one, I'd never know which one to play, because often either will do. To try and address this, I added the "always start at this position" checkbox in the adjust screen. If you need the first three moves for context instead of the first two, you can do it there. (Let me know if more explanation is needed)

Ah ok, that's nice; it's good to hear how you're using the site. Maybe the wrong moves threshold could be adjusted to take into account a large variety of possible moves, or some way to advance the position without a guess? Do groups stay in your queue till you remove them or is there a threshold where they get removed automatically?

daniel_the_smith wrote:

Abydos1 wrote:It'd be nice if the hints were the correct colors and board orientation.

Yes... I'd have to generate 16 times as many thumbnails to support that. I'll do that eventually, if I think my poor little server can handle it. (The server is currently running in a slice with 256mb of ram and a 10GB hard drive.)

Can't the image be rotated & flipped client side, maybe html 5? This seems to indicate at least flipping can be done with css, not sure about rotation.

daniel_the_smith wrote:At first I was like, wait, I do use jquery to remove it? Then I looked at the source and I was like, oh, right. Yes, my ajax request triggers a page reload. Well, I will fix this eventually.

Great work on the site, I can see it being very useful. Have you thought about including side positions and common reductions? That's probably a lot of work expanding the database though . I would love to see a good source for the non-corner josekis though, it seems very hard to find a good reference for them although eidogo does have some.

Abydos1 wrote:Ah ok, that's nice; it's good to hear how you're using the site. Maybe the wrong moves threshold could be adjusted to take into account a large variety of possible moves, or some way to advance the position without a guess? Do groups stay in your queue till you remove them or is there a threshold where they get removed automatically?

I'll think about maybe giving you more chances for the very first move, that might work. Currently, everything stays in your queue until you remove it. I've considered adding code to "retire" positions once you know them really well, but at that point they'll only appear once a month or so, so it might actually be useful review-- I haven't decided what to do yet, basically.

Abydos1 wrote:Can't the image be rotated & flipped client side, maybe html 5? This seems to indicate at least flipping can be done with css, not sure about rotation.

Nice idea, thanks-- css is not my specialty

Abydos1 wrote:Great work on the site, I can see it being very useful. Have you thought about including side positions and common reductions? That's probably a lot of work expanding the database though . I would love to see a good source for the non-corner josekis though, it seems very hard to find a good reference for them although eidogo does have some.

Actually I've been planning out the next version of my tool for building the joseki tree database-- I've figured out how to do a lot of it much better and I'm pretty excited, though I haven't started on it yet.

Planned:
* Corner josekis-- I have thought of a way to replace my joseki identification heuristics with something that is actually correct.
* Corner invasion josekis-- my current issues (I don't think anyone has noticed, but I have some) will be fixed by the above.
* Half/Full board "joseki".
* Side joseki.
* Center joseki-- I'm not even sure what that means or what I'll see, but I'm planning on doing them.
* Perhaps some basic filtering-- by date and/or strength of players. I don't think I have the HD space to filter the transparent stones. But I can filter the moves for sure.
* All of this would be addable to your queue, so you'd be able to learn side reduction/invasions, etcetera.

Yeah, so if people have more requests for browse 2.0, now is the time to make them

Could there be some sort of highlighting or shading of the area in question when you hover over the hint at the bottom of the board, so that you can have some idea exactly what is meant by "in the corner" or "in the center" or "on the side"?

I keep running into instances where I'm looking at a follow-up position, decide I would attack from one side or the other, then I look at the hint and it says "White to play on the bottom."

I think "on the bottom? Really?! Hmm, ok. It must be this squeeze from the second line... no, this one. no... huh. Attach and cross-cut?" Then when I've convinced the program I have no idea at all what I'm doing, it shows me a move on the sixth line, the one I thought of initially.

I understand that this sixth line move might be "on the bottom" when dividing the board into quadrants (er, octets?) but it has a HUGE disconnect in my mind from anything that could reasonably resemble that phrase. If the area that was being referred to by "bottom" or "lower-right corner" or whatever could be highlighted in a different color when you hovered over the hint, that would be an amazing improvement, as I could make sure I'm thinking about the same part of the board the program is.

ethanb wrote:Could there be some sort of highlighting or shading of the area in question when you hover over the hint at the bottom of the board, so that you can have some idea exactly what is meant by "in the corner" or "in the center" or "on the side"?

I keep running into instances where I'm looking at a follow-up position, decide I would attack from one side or the other, then I look at the hint and it says "White to play on the bottom."

I think "on the bottom? Really?! Hmm, ok. It must be this squeeze from the second line... no, this one. no... huh. Attach and cross-cut?" Then when I've convinced the program I have no idea at all what I'm doing, it shows me a move on the sixth line, the one I thought of initially.

I understand that this sixth line move might be "on the bottom" when dividing the board into quadrants (er, octets?) but it has a HUGE disconnect in my mind from anything that could reasonably resemble that phrase. If the area that was being referred to by "bottom" or "lower-right corner" or whatever could be highlighted in a different color when you hovered over the hint, that would be an amazing improvement, as I could make sure I'm thinking about the same part of the board the program is.

Good idea, thanks! That has happened to me before, also. It's kinda hard to give hints vague enough to not give it away but useful enough that it's actually a hint. I probably shouldn't describe a 6th line move as "on the bottom", but I also don't want people to know that "bottom" means, say, 3rd line or lower-- I want you to learn the board, not the hint. Yes, I may have incoherent preferences. :/

(I usually ignore the hint until I've gotten it wrong a couple times, then I'm like, oh wow, I'm not even on the right side...)

daniel_the_smith wrote:

ethanb wrote:Could there be some sort of highlighting or shading of the area in question when you hover over the hint at the bottom of the board, so that you can have some idea exactly what is meant by "in the corner" or "in the center" or "on the side"?

I keep running into instances where I'm looking at a follow-up position, decide I would attack from one side or the other, then I look at the hint and it says "White to play on the bottom."

I think "on the bottom? Really?! Hmm, ok. It must be this squeeze from the second line... no, this one. no... huh. Attach and cross-cut?" Then when I've convinced the program I have no idea at all what I'm doing, it shows me a move on the sixth line, the one I thought of initially.

I understand that this sixth line move might be "on the bottom" when dividing the board into quadrants (er, octets?) but it has a HUGE disconnect in my mind from anything that could reasonably resemble that phrase. If the area that was being referred to by "bottom" or "lower-right corner" or whatever could be highlighted in a different color when you hovered over the hint, that would be an amazing improvement, as I could make sure I'm thinking about the same part of the board the program is.

Good idea, thanks! That has happened to me before, also. It's kinda hard to give hints vague enough to not give it away but useful enough that it's actually a hint. I probably shouldn't describe a 6th line move as "on the bottom", but I also don't want people to know that "bottom" means, say, 3rd line or lower-- I want you to learn the board, not the hint. Yes, I may have incoherent preferences. :/

(I usually ignore the hint until I've gotten it wrong a couple times, then I'm like, oh wow, I'm not even on the right side...)

I'm afraid the only way to avoid that would be to analyze the position with GnuGo or something. Then (if you're lucky and it classifies the position the same way most people would) you could use that info to describe the move as "White to attack the group on the bottom" - which brings a much vaguer picture to my mind than "White to play on the bottom" does, and can include all sorts of possibilities.

Just jumpimg on the thread to thank daniel_the_smith for the amazing work on dailyjoseki.
I just love it, it makes me want to try things i would not have considered.
And yes,shading the area for next play would be great instead of a vague sentence.
Also a visual indication on who is to move , i keep forgetting to read the sentence on the bottom and when a side play twice because the other tenukis i feel stupid..
i guess a big colored stone should jump more to the eye

Suppose I start with my favorite, the 3-4 high approach, low pincer. There are many moves possible there--too many to have any idea. But here are three ways the tutor might handle it:

1) Work through them in order. Pick any reasonable move in the global context (where reasonable == moves that pros play) , play through the joseki, then return to the start and pick a new move. This won't work with moves that have a high branching factor, but I'm not sure the current approach works well there either.

2) The tutor could reject moves it wasn't looking for, without penalizing you. It could effectively say "that's a fine move, but not what I'm looking for."

3) A fusion of the first two: it could make you try and find all the reasonable next moves, then place the one it wants you to study on the board, then repeat.

None of these seem perfect, but I found the arbitrary choice of next move problem to be really hard to get past.

I just had this position occur while studying and it finished the sequence too early and I wasn't able to advance it any further along. Since I can't link to that position here's a picture of the sequence:


My username is the same if that helps.

Edit: After trying to adjust this position and going back to the study view it asked me how hard it was to remember and this was the first time I'd studied that sequence.

Also, I noticed I can't browse joseki in two separate tabs; not a huge deal but it's nice to have.

perceval wrote:Just jumpimg on the thread to thank daniel_the_smith for the amazing work on dailyjoseki.
I just love it, it makes me want to try things i would not have considered.
And yes,shading the area for next play would be great instead of a vague sentence.
Also a visual indication on who is to move , i keep forgetting to read the sentence on the bottom and when a side play twice because the other tenukis i feel stupid..
i guess a big colored stone should jump more to the eye

Thanks, I'm glad you like it! I do the same thing, I agree, I should have a colored stone instead of the word "black" or "white".

hyperpape wrote:Suppose I start with my favorite, the 3-4 high approach, low pincer. There are many moves possible there--too many to have any idea. But here are three ways the tutor might handle it:

1) Work through them in order. Pick any reasonable move in the global context (where reasonable == moves that pros play) , play through the joseki, then return to the start and pick a new move. This won't work with moves that have a high branching factor, but I'm not sure the current approach works well there either.

2) The tutor could reject moves it wasn't looking for, without penalizing you. It could effectively say "that's a fine move, but not what I'm looking for."

3) A fusion of the first two: it could make you try and find all the reasonable next moves, then place the one it wants you to study on the board, then repeat.

None of these seem perfect, but I found the arbitrary choice of next move problem to be really hard to get past.

Oh man I just thought of the perfect way to handle this. With my current data, I can have it stop penalizing you for moves that other pros made in the local position. That's cool, but will be too generous later on, when the whole point is to pick the variation that's correct for the whole board. But! With my next set of data, I can not penalize you as long as your move matches a full- or half-board position... ("That's a fine move, but in this game the pro did something different.") Thanks!

Abydos1 wrote:I just had this position occur while studying and it finished the sequence too early and I wasn't able to advance it any further along. Since I can't link to that position here's a picture of the sequence:

Hm, you can see from the 7/7 that it only found 7 moves to put in the test at all. I'm not sure why I'm generating test positions with only 7 moves. If you use the "remove test" option, it'll automatically give you a replacement. Also I can run a query to see which tests people reject with the reason "this test sucks"-- I think I wrote that, I'll double check later tonight.

Abydos1 wrote:My username is the same if that helps.

Edit: After trying to adjust this position and going back to the study view it asked me how hard it was to remember and this was the first time I'd studied that sequence.

I should probably word that "how easy will that be to recall?" the first time you see a test. ^^ (the reason you may not have noticed before is this may be the first time you passed a test the very first time you saw it!)

Abydos1 wrote:Also, I noticed I can't browse joseki in two separate tabs; not a huge deal but it's nice to have.

Yeah, I store state for that server side, not a great design! It was the first part of the site I wrote, eventually I will fix it.

daniel_the_smith wrote:

Abydos1 wrote:Edit: After trying to adjust this position and going back to the study view it asked me how hard it was to remember and this was the first time I'd studied that sequence.

I should probably word that "how easy will that be to recall?" the first time you see a test. ^^ (the reason you may not have noticed before is this may be the first time you passed a test the very first time you saw it!)

Ah yes, that would be it.

Recently I've noticed some weird lag, for example move attempts showing up in different orders. Are you using ajax queries for each move? If so, can't you just load the whole sequence client side?

Abydos1 wrote:Recently I've noticed some weird lag, for example move attempts showing up in different orders. Are you using ajax queries for each move? If so, can't you just load the whole sequence client side?

Gah, I've been waiting for someone to call me on that, you get the prize. You guys are finding everything now.

The best solution is to make the client smarter. Well, I have a temporary solution in mind that will not require writing a bunch of js. Also, my agreement with GoGoD includes me not giving out their games-- currently, if you look at how I do things, you'll see that it would not be possible to hack even a significant portion of a game, which is a property I need to keep.

Hm, actually I just thought of a solution that does everything I need (including not requiring me to implement the rules of go in js). Sigh, something else for the list... Did I mention there's just one of me and I have a day job?

daniel_the_smith wrote:

Abydos1 wrote:Recently I've noticed some weird lag, for example move attempts showing up in different orders. Are you using ajax queries for each move? If so, can't you just load the whole sequence client side?

Gah, I've been waiting for someone to call me on that, you get the prize. You guys are finding everything now.

The best solution is to make the client smarter. Well, I have a temporary solution in mind that will not require writing a bunch of js. Also, my agreement with GoGoD includes me not giving out their games-- currently, if you look at how I do things, you'll see that it would not be possible to hack even a significant portion of a game, which is a property I need to keep.

Hm, actually I just thought of a solution that does everything I need (including not requiring me to implement the rules of go in js). Sigh, something else for the list... Did I mention there's just one of me and I have a day job?

That makes sense not giving out the game. I'm not sure what your idea was but I'd think you could just store board states in an array where each index is similar to the return from each successive ajax query on a correct move; all you should need is the correct move and a list of stones to add.

How does it determine which test to give you next? Sometimes I've had the same sequence twice in a row. How is the EF adjusted each time? Does it take into account how long it's been since it's shown you that position?

Keep up the good work, I'm looking forward to your improvements

Click Here To Show Diagram Code

[go]$$Wcm1
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 0 1 7 . . . . . . . . . . . . . . |
$$ | . . 6 4 5 2 . . . , . . . . . X . . . |
$$ | . . X 3 8 . . . . . . . . . . . . . . |
$$ | . . . 9 . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . , . . . . . , O . . |
$$ | . . . . . . . . . . . . . . O . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Click Here To Show Diagram Code

[go]$$Wcm11
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 2 . 1 . . . . . . . . . . . . . . |
$$ | . . X O O . . 3 . . . . . . . . . . . |
$$ | . . X X O X . . . , . . . . . X . . . |
$$ | . . X O X . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . , . . . . . , O . . |
$$ | . . . . . . . . . . . . . . O . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Wow, W11 looks so strange; I've never seen it before. Neither eidogo nor josekipedia has it listed.

That is a weird one. Without checking, my guess is that this shows why automatically getting moves is better than hand-entering them. Perhaps it is from old games or something? It found 30 examples, so it's actually not that rare... Looks like a good move to confuse people with!

Protip: registered users can click the "embed a link" button from the browse page to get one of these:



I'll try and answer your other post later

daniel_the_smith wrote:That is a weird one. Without checking, my guess is that this shows why automatically getting moves is better than hand-entering them. Perhaps it is from old games or something? It found 30 examples, so it's actually not that rare... Looks like a good move to confuse people with!

Moyo Go's database shows dozens of results from the 70's through 2008, almost all of them are played with a white stone along the left side but I was surprised to see this game with nothing else in the top left.