Opening database

I've compiled a database of games between the current top 20 rated players and everyone with a rating of 1900 or more. Since some players have improved over time there are still probably some "bad" games in there but nonetheless they are mostly played well.

You can see the results here: http://matt.stumpnet.net/hexdb.html (this should work in most browsers, but I can't make guarantees)

For each move the board shows how many players played it and the percentage that won with that move. If you click on a move you can see the responses to that move and their percentages. If you keep clicking you can work your way through the possible games. If you end up on a branch with just one game it'll show you the link to the LG game.

Note that I've included the games in their rotated form as well, since they are equivalent. So you might end up in a game that's the rotated form of the actual game.

Thoughts?

When you say "won with that move," which player do you refer to if the move was swapped?

Swap is considered just another move by white. So the number you see next to "swap" is the percentage white won by swapping.

OK, so here's one easy observation: The most popular openings are indeed pretty even (i.e., winning percentage pretty close to 50%), so these players show good judgement picking them as opening moves.

Also: Note that this diagram is clickable! (I didn't realize that until just now). Click on a move to see percentages for follow-up moves. If you follow until there is just one game, that game will be identified.

Cool, good work HappyHippo!

By the way, the database seems to be outdated - there are only 13 games starting with c2, and none of them are mine, even though I've been playing it a lot recently.

Very interesting! I think E2 and K2 are remarkable. There should not be popular opening moves with win-percentages significantly above (or below) 50%. Agree?

Yes, I should have emphasized that it's clickable! While the first move stuff is interesting, it's been discussed a lot. The part I find the most interesting is the next few moves.

shalev - yeah, I think it's been a few months since I've updated. This began as a desktop program, I just recently converted it to a website. I should update it again, but the process isn't fully automated so it can be a bit tedious.

Tony - The interesting thing about those moves (especially K2) is that they only do so well because they are very commonly swapped, but are apparently quite weak (only 24% of people who swapped K2 went on to win). It seems players overestimate the strength of those moves? Players who didn't swap fared much better.

For the first move, the essential information is (wins by black when white does not swap + wins by white when white does swap) / total number of games.

Does the database exclude games which were decided by time-out/forfeit? There are many such games, and I think including them adds a lot of noise to the data.

Games in which a player times out are not included.

Quotes: For each move the board shows how many players played it and the percentage that won with that move. If you click on a move you can see the responses to that move and their percentages. If you keep clicking you can work your way through the possible games.
...
Swap is considered just another move by white. So the number you see next to swap is the percentage white won by swapping.
Endquotes

Okay, for example, for 1.A3, if I understand you correctly, out of 778 games with that first move, black won 50% or 339 games. Presumably this includes games where white swapped as well as games where white did not swap. Am I right so far?

When I click on A3, the swap circle off the board says 120 games, 43%. Does that mean white swapped 120 of the total 778 games? And white won 43% of those 120, which rounds off to 52 games, yes?

When I add up the totals of the games, it appears that games where white swapped are not included in your tree of variations. Is that correct?

I appreciate your hard work. You might be interested in a similar chart for Twixt , created by Alan Hensel. It has no tree of variations, but it does have three values for each first move, and it uses a color coding system. I like Alan’s chart because it shows me what could be called the intrinsically balanced nature of each move, independent of whether it was swapped or not.

Yeah that's basically it. You can click on that swap circle and see those games as well.

I've been considering adding an option to fold the swapped games into the unswapped ones, basically remove the swap move and mirror the whole game. Then you would see the "true" percentages for each opening.

How are you or others creating these databases? Is the LG games database available or are you scraping games?

When I created a similar database (see http://www.littlegolem.net/jsp/forum/topic2.jsp?forum=50&topic=554 and http://canyon23.net/hex_first_moves/ ) I scraped games; I don't know of a simple way to get the LG database. Be sure to scrape slowly and politely. You might also consider becoming an LG member if you put more than the usual load on the server.

Yeah that's what I had to do. The downloader I made had a delay of several seconds between downloads. The text files representing the games aren't that large anyway, only a few hundred bytes each.

Amazing work :) I would love to see something similar for dots and boxes.