Fun facts about zero-k.info/stats

In another thread http://zero-k.info/stats/ was posted with comment "nobody knows what the numbers mean."
It was a bit offtopic there so decided to post in new thread.
I do not mean to attack those made that or similiar pages, just share some findings.
Just a bit confusing when that page is used to show all kinds of stuff and "small inaccuracies" are ignored.

Stats are funny and interessting but also have to keep in mind what they show and how to interpret it..
For some time I was wondering some things about that page but was too lazy to look more closely. Maybe others felt similiar? No?
Then the following blabla is even more boring than it is anyway:

The title says "Accurate player minutes stats" but pretty sure is not that.
One might read that as "time this game was played" but better title seems to be "time this game was hosted in lobby", which is NOT the same.

boring fact #1
gather
http://code.google.com/p/zero-k/source/browse/trunk/other/NightWatch/Nightwatch.cs#128
draw
http://code.google.com/p/zero-k/source/browse/trunk#trunk%2Fother%2FNightWatch%2Fweb%2Fstats

boring fact #2
&ext=maps in url to see map stats instead, like
http://zero-k.info/stats/display.php?stamp=1392102941&ext=maps


fun fact #1
I made a new mod, opened a room in lobby and never started it:

Already over an hour of playtime \o/
(But sadly adding 40 NullAIs did not make it count 40x faster)

fun fact #2
But nobody would do that, just go host a game and then never play?
Good thing we have machines for that!


But the effect of empty autohosts can surely not be so large, can it?
Go to http://replays.springrts.com/search/ and search for "Evolution RTS" and see how many games were played on 10 Feb. 2014
Answer is: None.
last games were 8 Feb 2014:


Now compare what the stats page says about EvolutionRTS on 10 Feb. 2014:

8510 player minutes: Third most-played game on this day.
Without a single game actually played, just from autohosts. That is more player minutes than other games that were played on that day.

fun fact #3
An empty autohost generates ~1440 "playerminutes" per day. (circa, minus downtimes of Nightwatch and autohost)
To keep with EvolutionRTS as example, when players get together to play it is usually 4-6 players for 3 hours or so. How does that compare to the autohost-minutes?
Example from 6 Feb 2014 = 322 playerminutes
The stats for that day show: 10749 playerminutes.
So on this "active day" only 3% were real playerminutes.
(Over a whole month the percentage would be even lower.)

fun fact #4
zero-K has empty autohosts too.
Its number change* atm there are 13 zK empty hosts:

13 * 1440 = 18720 playerminutes, just from empty autohosts for zK.
Clicking around a bit the stats show 50000 to 90000 daily playerminutes for zK
Asuming 70000 as average: ~25% of recorded zK playerminutes are from empty autohosts.
(*if one player idles in a room waiting for opponents and a new empty springie is spawned, that is 3 accounts generating playerminutes for one idler)

fun fact #5
Whenever relay-hosting breaks, a certain game called "dummy" suddendly appears on stats:
http://springrts.com/phpbb/viewtopic.php?f=16&t=27281&p=507277#p507277
(and is more popular than actual games)

(Feel free to correct math fails)

TL;DR for people who didn' read [2up]knorke's post: ZK needs more autohosts! Honk!

I wonder whether "minutes/player" (and "new players") count specs too. If so, does Springie count as a spec?

Anyway, interesting stats! I was really confused as to how EvoRTS got these numbers when I took a closer look at these stats.

quote: Then the following blabla is even more boring than it is anyway

You could have made it less boring by concisely stating your point.

Oh, wait! You did! My bad. Here it is in case anyone else missed it:

quote: One might read that as "time this game was played" but better title seems to be "time this game was hosted in lobby", which is NOT the same.

I assume it would be fairly easy (and easier than doing a scrapper) for you to actually make a simple frontend to such a query and commit it to site proper.

Only downside is having to get hands dirty in .net mvc, but it's for the greater good?

quote: fix replay section

what's wrong with replay section?

Its well known that /stats count specs. People asked me to NOT change that in order to keep historical perspective intact (don't change way/what you measure in the middle).

Regarding ladders/games I don't see how it's useless.

I had some fun writing a scraper consuming output of replay index... but due to the way data is displayed ("5 months ago"), it is only possible to get 1 month resolution that way.

I think the only way to get around that is to iterate over all battle details and get the date/time from the replay download link. It is not a problem, as I really would like to see accurate stats. But it is beyond stupidly inefficient... and will generate about 235'000 requests (+ 3-5% due to rampant win32/sql errors) for a total of about 6.5GB, so if someone with db access want to provide data before I take that route, you are welcome.

Any other efficiency improvement would be welcome, such as displaying date of game in the index partial view or optionally returning data in json format. But no action is strictly required.

quote: People asked me to NOT change that in order to keep historical perspective intact (don't change way/what you measure in the middle).

I wonder what those people hope to read from the stats?
Maybe they were aware of some flaws did not know it was this much?
(Or maybe they are just really interessted in seeing which game runs the most autohosts)

I think without specs&autohosts and only counting ingame time it would be much more useful. Maybe it could keep tracking "the old way" but also start second improved stats?

quote: Regarding ladders/games I don't see how it's useless.

You posted this image, without any other text:

Imo there are many open questions to what it shows:

"unique players+specs":
-When are they counted: When logging into lobby? On joining a room? Playing ingame?
-Do specs include autohosts?
-What would it look like if players and specs were not clumped together?

"minutes/playes":
-Minutes doing what? Playing? Being logged into lobby?

"new players":
-Does this mean new accounts created? Or low-rank players doing...something? (play? spec? idle?)

And then: What did you want to tell with the graph?
How is it relevant to thread where it was posted?

From recorded stats is it possible to get something like "Numbers of real players playing ingame" stats? Or "number of games running."
Because I think that is what ultimately would be most useful.

Wow, going by unique players and games I started doing ZK casts 6 months later than I should have. What happened soon afterwards to cause such a drop until July 1st (when I started casts)? The intake rate seems fairly constant, so apparently there was a lot of attrition then.

I think that the rise and drop is seasonal.