Practice game: Dead, unsettled or alive?

I got to 12 in one of the runs, but it’s not a uniform score for me - one run might be good and then the next this one shows up

and I’m like “why isn’t it alive” - took me a while to figure it out (so at least two runs went badly when this showed up :stuck_out_tongue: )

6 Likes

Basic, rectangular 6 dead in corner. (Failure: get a ko btw).
Got 8 at first try, didn’t get really the challenge (so would have tried more blitzy if so) was surprised by the timeout.
Now i am so happy with the infinite as i can try all of them without having to answer the obvious ones again and again.

Some more suggestions:
Let strong players give us their high score and then fix a realistic time with some chance to do all of them.

Automatic shuffle of the ordering so that you have more fun to do it again.

That’s all.

The problems are automatically shuffled for every new game, is this what you want or something else?

1 Like

Perfect, yes! Sorry, was already made.

1 Like

What about broadening the class of shapes slightly, to also include various common “open formations”, where the eye space is not yet fully enclosed, such as L-group, L+1, L+2, J group, long L, tripod, hovercraft, with variations on potential internal stones, hanes, descents, etc.

3 Likes

Maybe that could be an option - like a choice to have just these internal shapes with all liberties filled or something more general.

One can also do some similar shapes as ones in the puzzles where there’s an outside liberty or so, and maybe a weak connection somewhere.

2 Likes

It would be cool to have a number of different categories (pure nakade, common corner shapes, capturing races, maybe positions where ko is involved, etc) and then the user could select one or multiple of these categories to be tested on.

I will focus on completing the nakade collection first, but then I’ll think about these expansions. If anyone wants to take it upon themselves to make one of these themed collections, let me know and I can share the tool I used to collect and organize the positions.

4 Likes

I really like this sort of approach towards presenting and curating tsumego.

Tsumego are often presented with a prompt giving one specific color to play and usually provides a big hint by stating the specific outcome (e.g., “Black to play and live”, “White to play and kill”). For tsumego framed in such a fashion, the solver is typically tested through interactively solving. This approach requires a lot of effort for the curator in transcribing several variations and demonstrative responses.

I like the style of “status” tsumego, since it better captures a real game situation, where one is not given a direct hint about the outcome of a group, and one usually needs to reconsider each position both ways, in order to figure out whether one should tenuki.

It’s also quite convenient that only a ternary question needs to be handled, avoiding the preparation and implementation of variations. The scoring system makes it disadvantageous to blindly guess, provided that the three classes are roughly balanced (which does not currently seem to be the case).

8 Likes

Indeed, there is a big bias towards dead positions currently. My starting point was a list of dead eye shapes, and then it was harder to come up with good alive ones. I just need to spend some more time on it, but having an approximately even split among the three categories is definitely the goal (and should be for future categories as well).

6 Likes

In order to maximize ones score I think it would be best to blindly guess shapes one can’t answer immediately, so I think the penalty for being wrong should be increased.

2 Likes

If there are 3 choices with equal probability and you get 1 point for a correct answer and -1 point for an incorrect one, then, if you blindly guess,

  • in 1/3 cases you get +1 point (correct answer at first try)
  • in 1/3 cases you get 0 point (correct answer at second try)
  • in 1/3 cases you get -1 point (correct answer at third try)
    so that the expected score is 0.
3 Likes

Previously I had the wrong-answer penalty set to -2 points, to punish blind guessing a bit more. However, this really makes it feel like you ought to reset immediately after getting something wrong, if you’re trying to get a good highscore. So to me it felt like the current system maybe gives a better playing experience. I’m not sure though, interested to hear some more opinions on this.

Another scoring system to consider would be: +1, +2, +3 etc for each successive correct answer, and -1, -2 for each successive incorrect answer. This still has the same “issue” where a single wrong answer makes you want to reset if you’re trying to maximize highscore - but perhaps this is not an issue at all, especially since the games are so short anyways.

4 Likes

Since you are also considering the incentives to discourage resetting, maybe use a penalty for quitting early applied to the next game? Are we in the realm of meta-mechanism design?

Currently, I guess the most effective strategy is to simply guess as quickly as possible. If I understood correctly, dead is most frequent, followed by unsettled. Thus, guessing dead first, then unsettled, and finally alive (if the earlier guesses fail), one would generate a positive expected score for each puzzle. If someone is able to iterate this method very quickly, they could probably yield higher scores than by just playing normally.

1 Like

That definitely feels to me like a step too far in terms of complexity, for this kind of game :slightly_smiling_face: I like to have a thin layer of “gamification” on top of what could otherwise just be a flashcard app, but if one gets too stuck down in the details of the game implementation (in terms of lining up all the incentives to prevent every kind of cheating etc), it can detract both from the learning process and from the ability to enjoy the game. (not to say this is always the case, but I think as a general rule of thumb simple is often better - and I’m only slightly biased by the fact that simple is much easier to implement :wink:)

Even with the current problematic problem distribution, I think it’s not so easy to get a good high score using this strategy. I think after some practice an “honest” player could average more than 1 point per second (it takes less than 1 second to classify a shape when you just know the answer without thinking). To beat this, the “cheating” player would have to react several times quicker to the correct/incorrect sound (or visual) cues. Not to say it’s not possible, but it would have to be quite a dedicated cheater!

This gives me another idea for a way to patch out this cheat though: enforce a small wait (like 1 second) before the next guess after an incorrect guess.

But all of this will be needless once I actually sit down and fix the problem distribution! (I should be doing that instead of writing long forum posts about hypotheticals, huh? :stuck_out_tongue:)

2 Likes

Who wants to cheat anyway?

4 Likes

To some extent, if it’s too easy to cheat, it can ruin the game a bit, even if you’re just competing with yourself.

Like let’s say there was no penalty at all for guessing incorrectly: then I would probably be too eager to guess as soon as I get a hard problem, and the score would become less meaningful to me as an improvement metric.

But as long as “cheating” would take some amount of dedicated work, I agree it’s not worth worrying about in this context!

3 Likes

I like +1 and -1, blindly guessing is already bad idea because theres only 33% change of getting ir right. Maybe lot more seki and ko shapes to balance it out? ^^

Would be really nice to have some skill tiers with the shapes, and if you manage to get something like 5 points it would give you harder set next, or easier one if you finish with 0 :wink:

3 Likes

My issue with the current system is, that guessing quickly is better than reading the position. So in some way it is encouraging guessing. But it’s not really a big issue, I just think it might be better to come up with something else.

1 Like

What do you think of an incorrect guess being an immediate game over? This would be more conducive to actually taking some time understanding the position you just failed, compared to the current system which encourages moving on as quickly as possible.

If there is still a fixed total time, there should still be some kind of penalty for getting the last problem wrong, to not encourage guessing just before running out of time.

Alternatively, the goal could be to solve a fixed number of problems in as short a time as possible. But I think I prefer the fixed time, to keep the game length the same regardless of player speed.

2 Likes

For the timed mode, I like the current system of allowing players to continue playing, since under time pressure, there is an aspect of time budgeting and risk taking.

For the infinite mode, it is already effectively like a “game over” when a guess is missed, since the streak is reset to 0, and that’s what tracked.

What do you think about the following tweak? A player is only allowed to guess twice for each puzzle (it automatically moves onto the next puzzle after two failed guesses).

This would effectively create three outcomes for each puzzle (besides timing out):

  1. Correct on first guess +1 points.
  2. Correct on second guess +0 points.
  3. Wrong on both guesses -2 points.

This would give uniform guessing a negative expectation (we should still balance the puzzle distribution). By the way, even if the list of puzzles are imbalanced, the sampling can be weighted to fix the imbalance.