Revisiting Automatch Time Settings: Data-Backed Proposal for New Automatch Settings on OGS

[PRHG]

In an old thread they considered recategorising live games as <= 9s:

Blitz games not counted properly

Thanks @shinuito for bringing this to my attention again, I reckon the best fix would be to actually make those blitz games, ie dropping the average move time down below 10s per move. So poll time, what should be the time increment / byo yomi time ? poll I’ll update both the site wide scheduled tournaments and the automatch settings accordingly.

In the same thread anoek mentioned it would be a hassle to change things in the backend so that 10s counted as live:

Blitz games not counted properly

Once upon a time in a poll long long ago 10s was voted as the cutoff, and changing that has a lot of implications that are unpleasant - so that’s why it’s not listed. The solution is to make the things we called blitz games actually blitz games I reckon.

[Chrispytoast]

@BHydden

AFAICT it would only double the load on server resources without doing anything to streamline the kinds of games users are playing.

Can you elaborate on this?
I would imagine the AutoMatch code to be the most load intensive to continually search for games, but I do not see how posting a game to the Custom Game list would double the resources used.

In my ideal scenario:

• AutoMatch would post a game to the custom game list as well as being in the queue.
• Since the only thing that should be missing from posting a normal custom game is the title, that can be something like “AutoMatch request”
• AutoMatching will not need to scan the custom games since all AutoMatch games are already in the queue
• Required time settings are removed. Preferred settings remain with OGS defaulting to Fischer time.
• Fischer time should use reasonable increments to determine full game time (2x main time + 200 / 300 * increment) — 5min + 10s increment = ~54-70mins for a full game.
• The ranking of the game type is what is used for matchmaking. Ie 3k 19x19 Blitz gets 3k Blitz games even when their server rank is 1d

[sir_husky_potato]

Generally, I agree with you, but we should probably ask for details on the implementation for automatch, because it is possible that it would quite hard to semi-merge it with custom games. There is a product question that do we really need this if automatch would work greatly, but this is another thing. I personally like the idea of creating clear boundaries between automatch games (ranked, live games) and custom games (correspondence plus anything else), but this is really out-of-scope. I would love to see just the minimal list of features implemented - rapid / rapid+live time settings, new UI, and some numbers on buttons to show the number of active players of your rank looking for a game right now. As someone highlighted, even changing the default time from byo-yomi to Fischer is out of scope.

[Ajillo]

I would love to see this happen. I would play a lot more

[anoek]

Byo-Yomi settings for a new Rapid category seems pretty well agreed upon at 5m+5x30s (or 5m+3x30s, which results in nearly identical game durations)

The Fisher counterpart eludes me a bit though. I normally reach for some data analysis to answer such questions, however even with thousands of samples our data set yields some counter intuitive results, such as 2m+20s games taking substantially less time than either 2m+10s or 2m+15s games, as well as 2m+15s games taking substantially more time than 5m+30s games.

Trying to extrapolate game estimated game durations by doing math on estimated number of moves * move time and whatnot doesn’t seem to yield results that are anywhere close to reality, so I don’t think that method is particularly useful in this case either.

So, I think I’m going to ignore all of that and just try to get a feel from y’all as to what some reasonable times for Fischer clocks would be.

Specifically, those that have a feel for playing with a fisher clock, what do you think would be appropriate for Blitz, Rapid, and Classic time settings?

[_KoBa]

@anoek

what do you think would be appropriate for Blitz, Rapid, and Classic time settings?

How about:
blitz - 30 s + 5s/move cap at 1 minute
rapid - 5 min + 10s/move cap at 10 min
classic - 10 min + 30s/move cap at 30 or 60 minutes?

[Samraku]

@anoek

Specifically, those that have a feel for playing with a fisher clock, what do you think would be appropriate for Blitz, Rapid, and Classic time settings?

I see 1 byo-yomi period of 10s/move as the prototype Blitz setting in Go, so I would propose 30s +10s/move (4h cap). This will be a little more relaxed than 10s byo-yomi, and I believe OGS will currently classify it as Live, so reducing the increment to +9s/move should cause OGS to correctly classify it as Blitz, and make up for the more flexible nature of Fischer Time.

I normally play 10s/move increment regardless of game speed, so I would propose 15m +10s/move (4h cap) for rapid (I mostly play 20m +10s/move and 10m +10s/move, so this suggested setting falls nicely in the middle for what I think is a fairly typical “rapid” pace in online chess

While I would personally play 40m +10s/move or 60m +10s/move if I wanted a more classical game, 10 second increment is shorter than most people want for something called “classical”, and “classical” online is generally shorter than “classical” irl, anyway, so I would suggest 20m +20s/move for classical

To summarize:
Blitz: 30s +9s/move (4h cap)
Rapid: 15m +10s/move (4h cap)
Classical: 20m + 20s/move (4h cap)

[Samraku]

@_KoBa

blitz - 30 s + 5s/move cap at 1 minute
rapid - 5 min + 10s/move cap at 10 min
classic - 10 min + 30s/move cap at 30 or 60 minutes?

You seem to have a very similar feeling of appropriate settings as I do

[qnpnpmqppnp]

@anoek

even with thousands of samples our data set yields some counter intuitive results

Should we conclude that Fischer leads to large variations in game length, while Byo-Yomi is more predictable?

[shinuito]

@Samraku

Classical: 20m + 20s/move (4h cap)

I’d say this over

@_KoBa

classic - 10 min + 30s/move cap at 30 or 60 minutes?

I’m not sure you even want people to be able to bank 30 or 60 mins time for a semi casual (rated but not a tournament) game. At least if one is aiming for a popular enough setting.

I can imagine 15s or 20s being more than enough, but could still probably lead to several hours of a game in length.

Probably the upper end of the time control on how long a game could take could be a deciding factor on whether you queue up for it, rather than just what the average length typically is.

[_KoBa]

@shinuito

I’m not sure you even want people to be able to bank 30 or 60 mins time for a semi casual (rated but not a tournament) game.

Yeah this is actually quite good point. I was thinking that the cap should be higher for classic than for rapid, but maybe 30 minutes is too much. Even if players can bank up time till they reach the cap and then use that time as they wish, its probably not very fun experience when opponent takes a 29 minute break in the middle of the game.

Maybe these would be better for rapid and classic:
rapid - 2.5 min + 10s/move cap at 5 min
classic - 5 min + 30s/move cap at 10 min

And if players want to play with timesettings which resemble RL tournament games, open and direct challenges are always an option.

[Samraku]

@_KoBa

rapid - 2.5 min + 10s/move cap at 5 min

I feel like that’s moving into Blitz territory, semi-Blitz at best

[teapoweredrobot]

@anoek

counter intuitive results, such as 2m+20s games taking substantially less time than either 2m+10s or 2m+15s games, as well as 2m+15s games taking substantially more time than 5m+30s games.

Two questions:

Is there a cap difference?

Is there data about pacing of moves throughout the game?

I would hypothesise that there is a psycological factor at play that once the increment goes under 20s people are starting to feel that the game is a bit blitzy and so they play faster and bank more time (but then play slower later). Whereas a 20 or 30s increment makes people feel they can just play at a slower pace throughout.

I’m not sure what this means but I’ll take some confirmation bias of my position (and limited experience of irl tourneys) that Fischer is better when the increment is shorter so I would make the increments something like 10s in each case and control the game length with main time.

Maybe something like:

30s+8s blitz
5m+10s rapid
15/20m+12s classic

I would probably prefer to have 10s increment is all cases but I guess people will rather expect to see differences between the categories and I have in mind the “10s is not blitz problem”

I’m not sure I’m bothered about caps. If people want long/unlimited caps then this should be possible as a custom game. For these cases the cap can just be main time.

Off topic slightly but I’m not loving the designation “classic”. What we really mean is “online normal” or something. Classic to me would be something 1hr per player upwards… (Edit: how about “casual”? Maybe confusing with ranked/unranked but in respect of time hopefully conveys something between “fast” and “slow/thinky”)

[anoek]

@qnpnpmqppnp

Should we conclude that Fischer leads to large variations in game length, while Byo-Yomi is more predictable?

Not really. For instance, with 28k and 29k samples respectively, 1h+5x30s had a median duration of 12:12 where as 10m + 3x30s had a median of 15:51, and 25th percentiles of 7:06 and 9:51, 75th as 24:19 and 23:53 … so definitely some head scratchers with byo-yomi too.

[teapoweredrobot]

@anoek

with 28k and 29k samples

This makes me wonder about variation with player strength. Do weaker pairs play faster whatever the settings? I feel like I did

[anoek]

@teapoweredrobot

Is there a cap difference?

Not notably, here’s the data

  ct   | width | time_control | initial_time | max_time | time_increment |      _25th      |      _50th      |      _75th      |      _90th      
-------+-------+--------------+--------------+----------+----------------+-----------------+-----------------+-----------------+-----------------
  1183 |    19 | fischer      | 120          | 240      | 20             | 00:07:45.00988  | 00:10:45.222325 | 00:13:04.065193 | 00:15:40.519484  
  1678 |    19 | fischer      | 120          | 120      | 10             | 00:10:16.130627 | 00:15:06.487992 | 00:20:35.427857 | 00:26:05.328667 
  1124 |    19 | fischer      | 60           | 120      | 15             | 00:05:36.408546 | 00:11:35.506061 | 00:20:25.096841 | 00:31:02.67109  

@teapoweredrobot

Is there data about pacing of moves throughout the game?

That’d be a bit of a chore to obtain

[BHydden]

Shouldn’t you be looking more at the max time end rather than the average? Surely a quick resign has little to do with time pressure.

I also suspect you’ll see a difference in between 1x main caps and max cap, it’s a shame there isn’t a “no cap” option to make comparison easier, but I suspect people who want that setting probably use 4 hour cap regardless of other time settings.

@Samraku

To summarize:
Blitz: 30s +9s/move (4h cap)
Rapid: 15m +10s/move (4h cap)
Classical: 20m + 20s/move (4h cap)

I agree with this proposal.

10m main for rapid could also be good… try to target that “whole game in half an hour” demographic.

[Samraku]

@BHydden

10m main for rapid could also be good… try to target that “whole game in half an hour” demographic.

Yeah, I considered proposing 10m +10s/move for rapid. It has the advantage that it’s a natural progression from 10m +10s/move for rapid to 20m +20s/move for classical

It also, and I’m just throwing this out there, suggests 5m +5s/move as a possible Blitz Time Control. Anyone want to try it with me to see if it feels like Blitz?

EDIT: I’m headed to OGS to try 5m +5s/move
EDIT: I played a game of 5m +5s/move and it felt good. I wish there was an option to set a higher cap than 5 minutes, though

[Samraku]

@teapoweredrobot

Off topic slightly but I’m not loving the designation “classic”. What we really mean is “online normal” or something. Classic to me would be something 1hr per player upwards… (Edit: how about “casual”? Maybe confusing with ranked/unranked but in respect of time hopefully conveys something between “fast” and “slow/thinky”)

At this point “classical” is the term used online for what irl would be on the board between rapid and classical, so I think it’s most in keeping with current use to use “Blitz”, “Rapid”, and “Classical”

[Samraku]

@Samraku

To summarize:
Blitz: 30s +9s/move (4h cap)
Rapid: 15m +10s/move (4h cap)
Classical: 20m + 20s/move (4h cap)

@BHydden

10m main for rapid could also be good… try to target that “whole game in half an hour” demographic.

@Samraku

Yeah, I considered proposing 10m +10s/move for rapid. It has the advantage that it’s a natural progression from 10m +10s/move for rapid to 20m +20s/move for classical

It also, and I’m just throwing this out there, suggests 5m +5s/move as a possible Blitz Time Control.

@Samraku

I played a game of 5m +5s/move and it felt good.

Updated Suggestion:
Blitz: 5m +5s/move
Rapid: 10m +10s/move
Classical: 20m +20s/move

Preferably caps would be at least a few minutes higher than the starting time, but it’s acceptable to have caps at the starting time for Blitz if OGS cannot be modified to allow higher caps in Blitz