Talk:Cornering

From HexWiki
Revision as of 05:36, 2 May 2023 by Demer (Talk | contribs)

Jump to: navigation, search

I'm not sure whether how Katahex plays on an empty board is relevant to this article. The sequences shown in the article (other than the Katahex one) show how Red can definitely play towards the center, regardless of what else is on the board (as long as the amount of space shown in each sequence is available). They are verified by the dfpn-solver. By contrast, the Katahex sequences are just something Katahex seems to prefer in certain situations; they are not verified, nor is the amount of space known. Perhaps these should go into a separate, more speculative subsection? Selinger (talk) 23:14, 1 May 2023 (UTC)

I only included these sequences (and discarded several other candidates) because I saw them in many different scenarios in KataHex self-play, on larger boards like 19×19 where the corner happens to be empty, and not just on an empty board. I have played around with KataHex for 200-300 hours by now (yes, that's hours a day on average), and I thought there were many places where the wiki's advice seemed outdated. There appears to be a big difference in philosophy; here's my side:

  • Currently, many articles only contain things that can be verifiably proven in a mathematical sense. That's okay, but I imagine many people come to the wiki only to try to improve their play.
  • As a specific example, I think Blue's move 12 in the first "d3 corner move" example is usually weak; in a real game you should probably play a bridge-move away from move 10. The problem is that while the example might be the provably optimal way to climb to "*", the goal of climbing to "*" doesn't seem particularly useful except in the specific situation where "*" is already connected to the top.
  • My goals are largely practically minded — I'm trying to contribute insights that I think are useful in real games (particularly larger boards where relevant advice is lacking in this wiki), where you're trying to reason under uncertainty (because you can't possibly calculate everything on a large board), so you need imperfect heuristics. It's not uncommon to see a totally empty corner in 19×19, where the KataHex sequences are relevant, and it doesn't matter exactly how much space is needed as long as you're fairly sure it's enough. I would argue that an empty corner is far more common than the situations where climbing to a particular hex at "*" is crucial. The KataHex sequences I added are good even if Blue has a ladder escape like in this position.
  • There are many other examples/articles (climbing) where the results hold theoretically, which is awesome, but I personally wouldn't recommend a player look at the scenarios if (s)he wants to get stronger, because "climbing to the nth row" is rarely a useful goal in itself unless you consider other factors. There are also many articles which seem highly outdated but might remain for historical reasons (V vs. H game 1; most strong players now would recognize 2...i1 as a very weak move because of Red j2).
  • Which leads to a meta-question: how high of a standard should there be for wiki contributions? I think most of my contributions have high practical value but low theoretical value.

Sorry for the long reply. If you think some of my contributions aren't additive, I won't be offended if you or others let me know. Hexanna (talk) 01:42, 2 May 2023 (UTC)

I agree with most of that! But perhaps we should draw a distinction between specific facts about tactics and general strategic advice. The articles you mention (like the one on climbing) are all about tactics, and I think there is value to knowing specific facts, like "such-and-such is a 4th row ladder escape", or "Red can get such-and-such switchback", or "Red can climb to the 7th row using this much space". To me, this is qualitatively different from general strategic advice like "in some situations, climbing might not be such a good idea" or "Katahex prefers another move to playing the switchback here". I'm not saying that such strategic advice isn't valuable; just that maybe it shouldn't be mixed together with tactics. Selinger (talk) 05:03, 2 May 2023 (UTC)

Hexanna, have you also seen those lines in self-play on ​ ​ ​ 11x11 , 13x13 , 14x14 ​ , ​ ​ ​ where 11x11 uses the c2 opening or the i10 opening or far-closer-to-50% 2-move openings? ​ ​ ​ ​ ​ ​ ​ My understanding is that those sizes are far more popular, and it could be that what you added only shows up significantly closer to 19x19, ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ Demer (talk) 02:32, 2 May 2023 (UTC)

The vast majority of my self-play experiments have been on size 15 and 19. I haven't seen the KataHex lines on 11×11, unsurprisingly considering the space it requires (and I've not run many games there). It's also rare on 13×13, because the bot consistently likes to take corners early in the game. The closest example I remember is something like this. I agree it shows up much more frequently on size 19, which is a common size on LittleGolem but perhaps less common in "fast" time controls. Hexanna (talk) 02:56, 2 May 2023 (UTC)


I have finally taken more than a passing glance at the long "D3 corner move" line's diagram: ​ Hexanna is of course correct about blue12 there. ​ ​ ​ For that move, anything towards the left other than bridging lets Red get at least switchback. ​ (Even if climbing to "*" was the actual goal, a switchback would suffice for that. ​ However, I think that really, Blue should first play Blue's forcing moves towards the right.) ​ ​ ​ I also agree with the rest of Hexanna's comments about climbing:

I think zippers are different-enough that that section should stay in the Climbing article, although some of the ​ 2nd row ladder ​ and ​ 3rd row ladder ​ Scenarios should probably be moved to a different article. ​ ​ ​ As one data-point for this, I note that I have used ​ Scenario 5 for 2nd row ​ (it was something like this) and ​ Scenario 3 for 3rd row ​ in games, although I don't remember ever using any of the later Scenarios in either of those two sections.

Demer (talk) 05:36, 2 May 2023 (UTC)