Arimaa Forum
        (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
      

        Arimaa >> General Discussion >> Server activity over time
        
(Message started by: Fritzlein on Dec 1st, 2004, 7:40pm)
Title: Server activity over time
Post by Fritzlein on Dec 1st, 2004, 7:40pm
I have created a graph of how much folks are playing Arimaa on the server over time.  I counted the number of games per four-week period, as well as the number of distinct players who played at least one game in the four-week period.  Unfortunately, my database is out of date, so the last four-week period is from October 7 to November 3.  All the great activity in the last month is not included.

http://www.math.umn.edu/~juhn0008/Activity_Graph.png

The good news is that already in October we were at an all-time high in the number of humans playing actively, the number of bots playing actively, and the number of games being played.

The bad news is that the number of human vs. human games has not increased significantly over time.  In fact, the average number of games an active human plays per month against another human is less than one.  (I plead guilty on this score.)  It's still very much a man vs. machine and machine vs. machine game.  (The majority of games are human vs. bot).
Title: Re: Server activity over time
Post by omar on Dec 5th, 2004, 6:24pm
Interesting graph. Now if the stock market could recover like that :-)

Naveed recently also mentioned to me that there are not enough human vs human games. We bounced some ideas around and thought that it might help to increase this if there was a ladder system and people could challenge each other to move up the ladder. Im considering this, but still need to think through the details of the policies for challenging and accepting or denying the challenge.
Title: Re: Server activity over time
Post by Fritzlein on Dec 26th, 2004, 3:14pm
I reformatted and updated the graph to include an additional 4-week period.  We shot up to an all-time high in bot vs. bot activity, maintained the high number of active humans (70 different humans played at least one game during the four weeks) and even saw an uptick in the number of human vs. human games.

Server activity hit an all-time low sometime around last June.  I sure hope we're not in for a similar falling off this year.

[Edit] Updated through late December.  There was a small dip in activity, hopefully not leading into a spring decline.
Title: Re: Server activity over time
Post by Fritzlein on Jan 31st, 2005, 7:47pm
Updated through the four-week period ending January 26.  We set a record for most humans playing at least one game, most human vs bot games, and almost set a record for most human vs human games.  And that's only finished games, so it doesn't count the 80 postal games.
Title: Re: Server activity over time
Post by Fritzlein on Mar 23rd, 2005, 4:35pm
When you smooth out the number of games per week into a nine-week moving average, it is clear that we are [EDIT: were] at an all-time high in human vs human activity, by a wide margin.

http://www.math.umn.edu/~juhn0008/9WeekAverage.png
Title: Re: Server activity over time
Post by Fritzlein on May 30th, 2005, 6:12pm
I updated the graph with another 10 weeks.  It's clear from even one month that the new Player of the Month contest rules are having an effect, as the number of human vs human games has soared even further, up to about 25 per week.  Let's hope the trend keeps up!

[edit] I updated another 3 months or so.  We hit a bit of a summer slump in June, but the Metafilter mention gave us a bounce, and I expect there will be a significant buildup in games of all types as we enter the winter tournament season.
Title: Re: Server activity over time
Post by Fritzlein on Jan 27th, 2006, 8:21pm
I've updated the graph above (and on the Wiki) through the last 9-week average I can calculate, i.e. December 22.  The bot vs. bot activity hit an all time high, thanks in large part to all of Gnobot's practice games.  Human vs. human games have dipped slightly since a peak right before the WC.

The most visible trend is that human vs. bot games are in freefall.  That type of game mostly comes from either bot-bashing record attempts, or from new players joining and working up the ladder.  Most folks rated over 2000 aren't contributing much to that stat at the moment.  Maybe that will spike a little if there is interest in playing the 2006 bots, or if we get some favorable press somewhere.

I think we've got a great little community as is, but I am nevertheless eager for more people to discover the joy of Arimaa.  More people means more fresh ideas and more fun for everyone as we figure out this crazy game.
Title: Re: Server activity over time
Post by Fritzlein on Mar 12th, 2006, 12:29pm
I've updated the graph again.  It looks like the bot ladder is doing wonders to increase HvB activity.  Indeed, since it is a 9-week average, there hasn't been time yet to show the full effect.  If we keep at our current rate, we will hit a new high of 350 or so HvB games per week.  Of course, that is contingent on new players continuing to join and stay around as the old players get bored with all the bots.  I'm thrilled how many fresh faces there are these days, and how many of them are actually sticking around for a while to progress through the ladder.

In other good news, all the automatic BvB games are not dominating the HvB games by any means.  At the current rate the average will rise to 100 or 150 games per week, and maybe less.  If the server load stays high due to lots of HvB games, perhaps fewer BvB games will even start.

The HvH games have ticked up slightly in spite of (or perhaps because of?) all the time folks are spending on postal games.
Title: Re: Server activity over time
Post by omar on Mar 15th, 2006, 6:05pm
It seems like there is always one or two live games going on just about every time I check. Also the number of people I see in the gameroom has definitely picked up.
Title: Re: Server activity over time
Post by Fritzlein on Apr 1st, 2006, 7:34pm
After pestering Janzert to use a log scale on his branching factor graphs, I realized that the server activity graph also should have a log scale.

http://www.math.umn.edu/~juhn0008/9WeekAverageLog.png

The substantial ups and downs of the HvH games were being drowned out on the linear scale.  Also the linear scale exaggerated the amount of dropoff in HvB activity in late 2005.  While it was a significant dip, it wasn't nearly as significant as the lull in mid 2004.

On a log scale, a straight line represents exponential growth.  It's pretty easy to lay a straight line along the top of the HvB activity, and predict when we will be up to 1000 games per week, i.e. mid to late 2007.  Won't that be fun!
Title: Re: Server activity over time
Post by Fritzlein on Apr 16th, 2006, 12:44pm
I updated the graph through April 15.  The HvB games are at an all-time high by a wide marign, and the HvH games are approaching their previous high.  The exponential growth continues!
Title: Re: Server activity over time
Post by IdahoEv on Apr 16th, 2006, 7:53pm

on 04/01/06 at 19:34:52, Fritzlein wrote:
we will be up to 1000 games per week, i.e. mid to late 2007.  Won't that be fun!


Not for the people who have to maintain the site, no.  ;-)
Title: Re: Server activity over time
Post by frostlad on Apr 17th, 2006, 2:10pm
Is there a way to look at the number of players that join over a period of time? It seems like there has been a large influx as of late.
Title: Re: Server activity over time
Post by Fritzlein on Apr 17th, 2006, 8:07pm

on 04/17/06 at 14:10:14, frostlad wrote:
Is there a way to look at the number of players that join over a period of time? It seems like there has been a large influx as of late.

I queried the games database for the earliest game time for each player, then grouped by month.  Here are the results through March 2006, since April is not yet complete:

[EDIT: image updated and moved down twelve posts]

The huge spike of newcomers in August 2005 was entirely a result of being mentioned on MetaFilter, but the corresponding spike in games played lasted only one week, not like the steady boom we have now.  Apparently, although the number of newcomers has been high recently, the present surge in HvB games is largely attributable to the bot ladder, rather than to better publicity.  The vast majority of the MetaFilter people played one or two games and left.  Nowdays newcomers seem to stick around for many more games.  The bot ladder is a challenge ("you can't do this") and an invitation ("you can do this") at the same time, a path laid out for all to follow.
Title: Re: Server activity over time
Post by 99of9 on Apr 17th, 2006, 11:20pm
It looks like something fundamental changed around Nov04-Feb05.
Title: Re: Server activity over time
Post by omar on May 5th, 2006, 9:20pm
In late October of 2004 Arimaa was mentioned on a popular Hungarian blog. But I don't that had a lasting impact.
Title: Re: Server activity over time
Post by chessandgo on May 9th, 2006, 6:45pm

on 04/17/06 at 23:20:56, 99of9 wrote:
It looks like something fundamental changed around Nov04-Feb05.



Maybe also in November02 ... :-X


(one day I'll shut up, I promise ...)
Title: Re: Server activity over time
Post by Fritzlein on Jul 5th, 2006, 8:55pm
I've updated through June 2006.  We've come down from our April peak, but less in the HvH category than in the other two, thanks in large part to chessandgo.  I find these days that I care about the red line most, and the blue line very little.  I expect the dip is part of a temporary cycle, and things will heat up again later in the year.  As the postal tourney starts to wrap up and people think ahead to the championship tournaments, activity in all types of games may hit new records.

[EDIT]Further updated graph appears below[/EDIT]
Title: Re: Server activity over time
Post by seanick on Jul 6th, 2006, 12:45am
I suggest that Nathan started the trend that Jean continued, though Nathan is apparently focused on Bot development and no longer adding to the HvH; Jean has remained active so I suppose that he is as responsible or moreso for the current human vs human boom.

I added my share of hvh last month though apparently just playing games isnt enough to be a potm, winning once in a while helps too :)

it seemed to me when I was first starting out on the bot ladder, that the way it was worded seemed to indicate you should not play any humans until you were done with the bot ladder. I realize now that the bot ladder is not intended to be played to completion before a human should be played, but I wonder if I was not the only one to get that impression. I don't remember what exactly did give me that impression and I don't see it now. maybe because its the only link in the play now area?

what if every online player accepting invitations showed up, not as a game table with decided rules, but just as a name/link which did the same as the "i" link next to their name in the online players list, but in the play now area? That might cause more newbies to blunder into a game, oops I mean play more hvh games earlier on. Maybe make the first level of the bot ladder a separate entity to the second and later levels so people at least have the basics down, and therefore don't get stuck in bot-only mode.
Title: Re: Server activity over time
Post by Fritzlein on Jul 6th, 2006, 8:50am

on 07/06/06 at 00:45:50, seanick wrote:
I suggest that Nathan started the trend that Jean continued, though Nathan is apparently focused on Bot development and no longer adding to the HvH; Jean has remained active so I suppose that he is as responsible or moreso for the current human vs human boom.

I added my share of hvh last month

Sure, I should have spread the credit around.  The top players of human games in May and June were

83      chessandgo
60      seanick
34      Fritzlein
31      Swynndla
29      Adanac
28      PMertens

Whereas in March and April they were

81      Swynndla
45      PMertens
42      OLTI
41      Ryan_Cable
34      chessandgo
26      Fritzlein
25      naveed


Quote:
what if every online player accepting invitations showed up, not as a game table with decided rules, but just as a name/link which did the same as the "i" link next to their name in the online players list, but in the play now area?

Hey, I love this idea!  The meaning of the little "i" flag is probably not obvious to newcomers, but if each player open for invites had their name as a link that opened the invitation box, newcomers would definitely feel more encouraged to play against humans.  I hope Omar finishes his new client soon so that he can add this feature.
Title: Re: Server activity over time
Post by chessandgo on Jul 6th, 2006, 10:51am
I like a lot this idea as well !
Title: Re: Server activity over time
Post by seanick on Jul 28th, 2006, 12:27am
given the neck-neck-neck-neck-arm-head-hat-hat- etc closeness of this months' player of the month I think the overall games hvh, and hvb have probably made an impact on the trendlines- where do you find the info again fritz? maybe it is my turn to start figuring these stats out.. in 2 or 3 days, if I haven't shown up to try to retain my meager 3 point lead in the POTM (which will probably last about another half hour as it is), you can bet by the end of the month I will have an in depth statistical analyzer of recent games instead of the PotM title  ::)

always the best priorities, my priorities.. like thinking about how to optimize my screen saver at work, or playing arimaa instead of going out and being social, saving $.05 by going to a specific store and getting a box of cereal on sale, but having to pay $1.50 more to get the milk while I am there...
Title: Re: Server activity over time
Post by Fritzlein on Jul 28th, 2006, 8:57am

on 07/28/06 at 00:27:00, seanick wrote:
always the best priorities, my priorities..

Don't underestimate the value in doing what grabs your interest at the moment.  I have heard that programmers are many times more productive when working on something fun to them.  So even if their energy is misdirected in one sense, it is an effecient allocation of resources in another.

Come to think of it, that's sort of like me playing Arimaa.  There are other projects I am always intending to do, because they would be "good for me".  I neglect these projects in order to play Arimaa.  Maybe this is costly to me in some sense, but in another sense, by doing something that engages me, I'm being as productive as I can be.

I will update the activity stats more often if there seems to be interest, although I don't want to forestall you from producing interesting summaries yourself.  My source is downloading the game databases that Omar updates on the weekend.  In order to give semi-current info, I usually try to do updates early in the week.

I like the 9-week-moving average stat because it smooths things out enough to see longer-term trends.  There have been brief spikes and dips in the past two months, but basically we've been holding steady at about 30 HvH games per week.  That has definitely picked up over the last two weeks, but it's still early to make a trend out of it.

On the other hand, I'm curious to see how you would slice and dice the data is some other way.  I don't pay any attention to longer vs. shorter, or rated vs. unrated, time of day, day of week, etc.  Please do any analysis you feel moved to undertake, for the education of us all.
Title: Re: Server activity over time
Post by Fritzlein on Jul 30th, 2006, 10:14am
Updated activity through July 2006.  The last few weeks have definitely seen more HvH activity.  Indeed, we're at a record-high 9-week average of 35 HvH games per week, but it's too early to know if that upward trend will continue.  I hope it does.  Perhaps the number of players who are bored with playing any bot has reached a critical mass.  The future may see the HvB line as a measure of the number of newcomers working up the ladder, while the HvH line measures the number of active established players.

[EDIT: Updated graph appears below]
Title: Re: Server activity over time
Post by Fritzlein on Dec 31st, 2006, 9:48pm
Apparently I hadn't updated the activity graph for the past five months.  Not that you missed anything, because it has pretty much been a flat line for the entire second half of 2006.  (See previous post for updated graph.) We didn't have a October-December slump like there was in 2005.  Will we have a February-April spike at the beginning of 2007 like there was at the beginning of 2006?  Time will tell.

If there is no seasonal surge in early 2007, I think we can safely say the exponential growth has petered out.  We're in a plateau of about 400 server games per week, not counting the automated BvB games.  When I was extrapolating exponential growth earlier in this thread I predicted 1000 games per week by mid-to-late 2007, but that doesn't appear so likely any more.

Here's a table of the number of games of each type by year:
HvHHvBBvB
20032813763671
200422547171075
200587610393944
20061384181372914

While I was at it, I updated the graph of the number of players joining per month.  The spike for December is at least partly attributable to new bot accounts.  Basically the news is that players didn't join any faster in 2006 than they did in 2005, but our number of games played increased significantly, so either newbies are playing more games each, or we are starting to retain more players.

http://www.math.umn.edu/~juhn0008/NewPlayersByMonth.png
Title: Re: Server activity over time
Post by Fritzlein on Jan 1st, 2007, 3:55pm
Woh was interested to know how many active players there are now compared to the past, as opposed to just newcomers.  An "active" player is defined as a human who has played at least one game in that month, either rated or unrated, either against a human or a bot.  Here's the graph:

http://www.math.umn.edu/~juhn0008/ActivePlayersByMonth.png

Further evidence that we are growing, but growing more slowly than before.  Incidentally, this graph isn't directly comparable to the one in the previous post, because the graph of newcomers counts bots as well as humans, whereas the graph of active players counts humans only.  For comparison, 58 bots and 140 humans were active in December.
Title: Re: Server activity over time
Post by woh on Jan 1st, 2007, 5:27pm
Thanks for the graph, Fritzlein.

It is great to see that the number of active players is growing.  The more players one can play against, the more interesting it becomes.
Title: Re: Server activity over time
Post by Fritzlein on Apr 2nd, 2007, 7:53pm
Well, it seemed like the gameroom had quieted down a bit recently, and the numbers confirm it.  I don't care about the BvB games disappearing, but it is sad both the HvH and HvB games are tailing off somewhat.  Not that we are in any danger of Arimaa dying out, but I sure would like to get mentioned on the Slashdot front page some day.

[EDIT]graph moved to last post[/EDIT]

I guess I can take my share of the blame for the HvH decline.  Maybe after my wife and I move into our new house and settle down I'll have a bit more time.
Title: Re: Server activity over time
Post by PMertens on Apr 10th, 2007, 5:39pm
I am certain that would look worse on a linear scale ...
Title: Re: Server activity over time
Post by Fritzlein on Aug 2nd, 2007, 8:05pm
Today Omar proposed a new way of looking at the growth of Arimaa.

The number of games played has been flat for about a year, and is well below our peak from March 2006, but this statistic depends a great deal on a small number of people playing lots of games (e.g. Arimaanator or syed).  Another statistic, for which I posted a graph above, is the number of unique humans playing at least one game in any given month.  However, this number fluctuates a great deal depending on the newcomers who play a game or three and then never come back.  Witness the Metafilter surge that caused our most active month ever by that measure.

Omar proposed instead measuring "repeat customers", on the theory that we have a larger and larger pool of folks hanging around.  I was skeptical that even our pool of repeat customers was growing, so we agreed on a new measure of activity that counts people but also weeds out folks who come and then leave.  I queried for the number of distinct people each month who played at least one game that was at least 30 days after their first game.  In other words, you don't start showing up in this statistic until 30 days after your first game.  The results:

http://www.math.umn.edu/~juhn0008/RepeatPlayersByMonth.png

By this measure our all-time peak was still March 2006, but at least we matched it in December 2006, presumably because of the World Championship and related events going on at that time.  So I was right that the pool of active players has not grown, probably Omar is right that the pool of lurkers is growing.  If he someday again has time for more events and promotions, folks will probably come back out of the woodwork to participate.
Title: Re: Server activity over time
Post by omar on Aug 4th, 2007, 12:50pm
Thanks for generating this Karl. It looks like we are about due for another surge here soon :-).

Omar
Title: Re: Server activity over time
Post by Fritzlein on Aug 12th, 2007, 10:18pm
Well, the surge from Reddit wasn't as big as from Metafilter, but it has been more sustained.  Perhaps that is because in the mean time Omar has added the bot ladder, which gets people hooked more easily.

http://www.math.umn.edu/~juhn0008/9WeekAverageLog.png

Actually, since this is a nine-week trailing average, it doesn't yet show the full effect of the bounce, and won't show it unless the increased activity continues.  If it does continue, we'll wind up near our April '06 peak of activity.
Title: Re: Server activity over time
Post by NIC1138 on Aug 13th, 2007, 8:26pm
Walking-means are ugly. >:( Use a modulated sinc filter, or an exponential decay recursive filter!!...  ::)

When was the reddit singularity?...
Title: Re: Server activity over time
Post by Fritzlein on Aug 15th, 2007, 9:12am
A less broad measure of activity is the number of points it takes to win Player of the Month as time goes by.

http://www.math.umn.edu/~juhn0008/PotMpoints.png

It's been tame in 2007 compared to 2006, but still a bit better than 2005.  In 2004 the PotM had different scoring that merely encouraged bot-bashing, so that's not included in the graph.
Title: Re: Server activity over time
Post by Fritzlein on Aug 17th, 2007, 2:04pm

on 08/13/07 at 20:26:47, NIC1138 wrote:
Walking-means are ugly. >:( Use a modulated sinc filter, or an exponential decay recursive filter!!...  ::)

I'm not familiar with these terms.  Doesn't "sync" imply matching something to something else?  If so, what are we trying to keep our signal in sync with?  Also, doesn't filter imply that we want to remove some sort of noise from our signal?  If so, what is the noise?

Maybe I should start at the other end.  What is ugly about a walking mean?  One property of the walking mean that I like is that it has the same integral as the signal.  It smooths things out, but doesn't make anything disappear.  You can see the MetaFilter event as a slight discontinuous elevation around August 05.  That's a compromise between showing a spike and editing out the spike completely.


Quote:
When was the reddit singularity?...

The post to Reddit was some time on August 3, and the surge of new players occurred over the following two or three days.

Interestingly, however, there had been a significant increase in the number of games played the week prior to the Reddit mention, which is something I can't explain.  Maybe it was just a convergence of unrelated factors.
Title: Re: Server activity over time
Post by NIC1138 on Aug 18th, 2007, 4:29pm
Sorry for the gross mistake, it's actually "sInc". It's the nickname of the sin(x)/x function :)
http://en.wikipedia.org/wiki/Sinc_function

In the point of view of linear filters,
http://en.wikipedia.org/wiki/Filter_design
the walking mean is equivalent to a very bad-looking filter that is actually a sinc funtion in the frequency domain. :P A sinc in time results in a "box" in frequency, what should be theoretically more beautiful. ::) This box can be made smoother using a pair of modulated sincs.

Now this is the theory for finite impulse response filters. But there are quite a lot of nice IIR filters that work beautifully for smoothing your data.

http://en.wikipedia.org/wiki/Iir_filter

Simply calculating y[n] = x[n] + y[n-1] can do a very good job in integrating your data. Now, you can have a "leaky integrator" simply doing something like  y[n] = x[n] + 0.25*y[n-1].

Any of these are much more beautiful then a walking mean!... I would advise at least using a Hamming window instead of just summing up N consecutive values with the same weight...

http://en.wikipedia.org/wiki/Hamming_window
Title: Re: Server activity over time
Post by Fritzlein on Aug 23rd, 2007, 3:09pm
Thanks for the explanation and the instructive links.  I don't know anything about signal processing, and it is fascinating reading.

I guess what I really want is a low-pass recursive filter such that the integral of the output signal equal the integral of the input signal.  I'll experiment with some values and post a new graph soon, perhaps Sunday when the latest week's numbers come out.

Thanks again for the suggestions and the interesting reading.
Title: Re: Server activity over time
Post by Fritzlein on Aug 26th, 2007, 12:56pm

on 08/18/07 at 16:29:59, NIC1138 wrote:
Now, you can have a "leaky integrator" simply doing something like  y[n] = x[n] + 0.25*y[n-1].

I decided to go with the low-pass (exponential decay) filter of y[n] = 0.25x[n] + 0.75*y[n-1].  To make it time-symmetrical I ran it both forward and backward, averaging the two.  The result is:

http://www.math.umn.edu/~juhn0008/IIRAverageLog.png

Aesthetically I am not sure it is more beautiful than the walking mean, but I do like other properties.  For example, you can now see the exact week in 2005 when the MetaFilter spike occurred, even though it is mostly smoothed out.   Also you can see that there was no Reddit spike, rather a broader surge over the last month.

Over the past month our HvB games have been almost as high as our peak of April 2006, but our HvH games are still trailing our peak of July 2006.


Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1!
YaBB © 2000-2003. All Rights Reserved.