OGS rank histogram (outdated)

Lys · August 29, 2018, 12:33pm

Moves are available: each one of them, game by game.
You “only” have to count them.

DVbS78rkR7NVe · August 29, 2018, 12:34pm

Nah, asking for moves is too much work for server so it’s only if you’re going to load games one by one. And one by one is very slow.

Here’s an interesting problem. If we want to analyze 19x19/live games, then ratings are also going to be in category 19x19/live. And they aren’t the same as overall rank, right. That’s a bit complicating.

smurph · August 29, 2018, 12:38pm

Well, if each finished game gets a heads-up #moves stamped onto the game, the API could simply add it as a column. Obviously if it had to do a separate lookup for every game it would be slower.

And yes, overall rank can, but usually doesn’t differ much from 19x rating (unless people mostly play other board sizes). My overall rating is 2371, overall 19x19 rating 2371, live 19x19 rating 2275. A quick look at someone who plays a good number of 19x live games gives 1611/1476/1459. The general impressian I get is that overall/overall rating is inflated by about 100 pts. To compare, someone I know who plays almost exclusively 9x9 sports 1661/1241/1558 (overall 19x dumpstered by terrible blitz rating). A second example shows the opposite trend: due to very few games, his blitz ratings are extremely inflated (1908) when the rest looks normal with 1654/1570/1597.

I would certainly not count blitz games, as live 19x19 seems to be the best predictor of overall rank.

That said, naturally, good data is hard to get.

DVbS78rkR7NVe · September 2, 2018, 12:36am

Sooo @smurph
19x19, live, ranked, no handi, no timeouts, stable (dev < 125) rating in live and overall. Most popular bots are excluded too. Everything seems to match up. I still used overall rank to calculate difference. It’s still not 100%. Maybe there’re just too many people who don’t care about their games?

Rank diff	Total	W wins	B wins	Ratio white/total
-9	46	4	42	0.09
-8	77	5	72	0.06
-7	91	6	85	0.07
-6	118	15	103	0.13
-5	163	19	144	0.12
-4	223	37	186	0.17
-3	322	80	242	0.25
-2	526	187	339	0.36
-1	1418	598	820	0.42
0	2673	1354	1319	0.51
1	3567	2089	1478	0.59
2	3763	2493	1270	0.66
3	3067	2266	801	0.74
4	2388	1958	430	0.82
5	1679	1442	237	0.86
6	1337	1206	131	0.90
7	1050	966	84	0.92
8	813	759	54	0.93
9	492	475	17	0.97

8-9 ranks upsets

Eugene · September 2, 2018, 12:50am

I had a look at the first one of these.

It wasn’t that the 2k who was beaten didn’t care. That player was actually beaten.

EDIT: actually, it appears to me that they were - it wasn’t “resign because the other person is too boring”. It might be above my grade to say that they were actually beaten…

And the person making the upset, who was at 10k at the time of the upset, went on to rank up to 2k in the next two months.

Then in the two months after that they went back down to 7k. All that time OGS thought that their rank assessment was approx +/- 1.5.

Seems to add weight to the claim in the other thread that OGS ranks are too unstable.

GaJ

meili_yinhua · September 2, 2018, 7:33pm

I’m pretty sure the OGS Automatic color algorithm doesn’t do this anymore…

I’ve been against weaker players as black in automatic (non-handicap) and the other way around…

I’m kinda wondering where this idea comes from in non-handi games, especially since bots are saying w w/ komi is better by a small margin.

DVbS78rkR7NVe · September 2, 2018, 8:16pm

“Anymore” doesn’t matter since I use older games also.

Idea of stronger player playing white comes from history, where else, right? There was no komi, so black had advantage of first move, in line with that master played white. And disciple played in upper-right corner, bowing to master, and allowing master to take upper-left corner without reaching too far.

DVbS78rkR7NVe · September 20, 2018, 7:59am

Accurate day-by-day graph of new users registering to ogs. Restricted to users with at least one ranked game because a lot (most?) of accounts are registered and forgotten without any activity on them.

AdamR · September 20, 2018, 8:17am

So the biggest inflood of users were not when Alphago started it’s reign and marked the new age of neural networks, but when Lee Sedol beat it that one time? very interesting.

I guess it was just that it was the most medialized part, but still

smurph · September 20, 2018, 11:03am

That’s nice, can you superimpose the # of users with <20 games and a rank >3d?

DVbS78rkR7NVe · September 20, 2018, 8:12pm

I didn’t quite get what exactly do you want, and what can we even get from this information.

3d+ is an elite club. By my count 2154 players have 3d and higher rating. And 882 of them have fewer than 20 (ranked) games.

BHydden · September 20, 2018, 8:28pm

Pretty much the only complaint I hear about this server on the internet is that not enough high dan players play here. If we can see specific times that a large number of them decided to join then that is a step in the direction of attracting more and boosting our international server rep. though with such small numbers it might need a separate graph rather than an overlay.

smurph · September 20, 2018, 8:55pm

I’m wondering if we see any spikes in the history of graphs for the ratio of [new3d+ / newSignups] for players with <20 games. An answer to the question “did the ratio of new dan players to new signups change at any point in time and if so, in what direction and by how much”.

Of course it would not matter much if this was extended to 1d+ or 2d+, my main question would be if the influx of new (1 month old?) dan-level accounts ever spiked when compared to the total influx of new accounts in that same period of time.

Fewer than 20 games means they didn’t steadily grow to be a dan player but they were before they created the account. Since the influx of many new players would at best lower the ratio of experienced to unexperienced players, it would be interesting to see if we see exactly that reflected in the numbers. Of course I don’t know if that’s possible to specify (account age), I’m just saying it could tell us something.

bugcat · September 21, 2018, 11:21pm

I wonder what happened on 3 Oct. 2013 that began the rise in base rate. Perhaps the merge with Nova.gs? I’m not sure when that happened.

Edit: Checked on SL, it says “the merge was announced on October 19, 2013” so it’s the right sort of time if a little later than the waypoint we see on the graph.

DVbS78rkR7NVe · September 22, 2018, 5:34am

Since merge with nova.gs OGS started to support live games. Correspondence levels mostly stayed the same. So all these players came for live games.

Kosh · April 27, 2020, 8:57am

Could I persuade you to redo the outdated OP histogram to better inform this discussion: Recognising and fully implementing ranks beyond 25 kyu?

Please.

Preferably with green covering deviation <160 as this marks the removal of ‘?’ rank and the humble rank effect on matchmaking (I think).

flovo · April 27, 2020, 11:30am

There is a more recent version at of the histogram at: