Online game data XML access
Online game data XML access
There is a dynamic API to get recent games as they come in. It will only return games after 300,000 right now.
It will return a maximum of 5,000 games at a time.
Mode 1:
XML for the last N games played. Minimum of 10 and maximum of 5000. Example URL:
http://sillysoft.net/lux/xml/gameHistor ... tGames=100
Mode 2:
XML for games since given game ID (above 600,000). Example URL:
http://sillysoft.net/lux/xml/gameHistor ... ame=605000
Mode 3:
XML for games between a certain ID range (above 600,000). Example URL:
http://sillysoft.net/lux/xml/gameHistor ... ame=600200
Cache the data you get back please. Don't make lots of requests for large sets.
If you want all the card/continent options for every game add &includeOptions to the URL. It all comes as one text string.
The seed given is always according to the current standing, not the historical standing from the game date.
Happy hacking.
- n00less cluebie
- Lux Cantor
- Posts: 8377
- Joined: Sun Jan 06, 2008 8:55 am
- Location: At the Official Clown Reference Librarian Desk--'All the answers you weren't looking for.'
- Contact:
Re: Game data XML access
Now we can all implement our own Ranking systems, and we can see what works and what doesn't.
Thanks!
- The Wontrob
- Ninja Doughboy
- Posts: 2792
- Joined: Wed Oct 03, 2007 9:56 pm
- Location: The Pan-Holy Church, frollicking
- Big Will E Style
- RAW Dogger
- Posts: 2943
- Joined: Tue Oct 24, 2006 1:28 am
- Location: Los Angeles, California
- Kain Mercenary
- Luxer
- Posts: 201
- Joined: Mon May 23, 2005 4:21 pm
- Location: OMEGA HQ
You can get wins and most common maps from the backlog of game information that Dustin provided. Alternatively, you can scrape the site, but I don't know how Dustin feels about that...Dominator wrote:Is there any way to use data from a player's ranking page such as wins, most common maps, awards, ect.??
- The Wontrob
- Ninja Doughboy
- Posts: 2792
- Joined: Wed Oct 03, 2007 9:56 pm
- Location: The Pan-Holy Church, frollicking
The page is also accessible from the wiki front page http://sillysoft.net/wiki/ (last link on the page).
I'll keep it up do date for a few weeks.
Cool! Greg, it would be fun if you could post your results on the Alternative Scoring System page.GregM wrote:I have an outline of a basic ranking program running that produces results vaguely similar to the official rankings.
- Scad
- Lux Elder
- Posts: 2521
- Joined: Sun Aug 13, 2006 6:53 am
- Location: Walking through the woods on a snowy evening
Also, I wonder if there are some interesting patterns regarding cumulative weekly scores over the long term... something like seeding, maybe? Or perhaps just an avg score/ per week stat
My system is just a quick hack, an experiment to see if zero-sum makes sense. It would need further tweaking to make it a "production ready" system. It does have it's flaws. Someone could win by stopping playing after a lucky streak, for example.Scad wrote:Bertrand, a problem I see is that with this less differentiated system, there's a greater possibility of ties. Raw usually doesn't because the difference is pretty significant, but with much smaller numbers the ties will likely go up. How would your system solve this?
Dustin's current system has the advantage of being "slightly positive" , rewarding the frequent player and preventing ties. It's a good setup, better than my system I think.
Good idea, I've put up a long term link here: http://sillysoft.net/wiki/?Long%20termScad wrote:Also, I wonder if there are some interesting patterns regarding cumulative weekly scores over the long term... something like seeding, maybe? Or perhaps just an avg score/ per week stat
ALSO, someone pointed out that applying new equations to old data could be somewhat irrelevant. Players change the way they play or who they play with based on the ranking system in place. For example, when dustin's RAW system was in place everyone played more games because there was less risk in losing.
Another example is if you re-calculated last years NFL season. Imagine touchdowns worth 8 and field goals worth 4. The decisions that the coaches made during the season would not matter now.
Using the XML data can show that your equation works in theory, however everything can change when it is actually put into practice. Even the tiniest loophole could be taken advantage of... Alias use for example.
Very true, I think it's Para who pointed this out. I agree that the results do not mean much.Dominator wrote: ALSO, someone pointed out that applying new equations to old data could be somewhat irrelevant. Players change the way they play or who they play with based on the ranking system in place.
But having the game data on-line is still infinitely better than what we had before. It enables us to find the obvious problems in our new systems. Dustin's last week "experimental disaster" could have been avoided by running the new formula against the historical database.
If we ignore the above (because there's really no way to compensate for it) how about this as an objective method of evaluating ranking systems: the ranking of players at the beginning of a game should predict the outcome of that game as often as possible.Dominator wrote:ALSO, someone pointed out that applying new equations to old data could be somewhat irrelevant. Players change the way they play or who they play with based on the ranking system in place. For example, when dustin's RAW system was in place everyone played more games because there was less risk in losing.
Another example is if you re-calculated last years NFL season. Imagine touchdowns worth 8 and field goals worth 4. The decisions that the coaches made during the season would not matter now.
Perhaps the following measurement would be good: if A is ranked above B, A should place higher than B in a game. The best ranking system is the one with the most successes in making this kind of prediction.
Random rankings give, of course, 50% accuracy. I ran a little test on games #500000-600000 and dustin's raw system seems to have a success rate of 59% -- that is, someone with higher raw will do better than someone with lower raw 59% of the time. If you look at seeds instead of raw (using raw to compare unseeded players) the success rate goes up to 61%. Not bad, given the unpredictability of Lux. But surely it can be improved.
- kitty on catnip
- Lux Elder
- Posts: 2209
- Joined: Tue Jun 06, 2006 12:34 pm
- Location: BACK IN THE FORUMS...
- Contact:
so, if you'd compare a calculation based on skill, that skill calculation that is above a lesser calculation should hypothetically be winning better than 61%...I completely agree here. I think 70-75% would be a much better goal to strive for. Any higher than that is shooting too high, for even the best players only win 1 out of 3 games. or slightly higher...
any ideas, you mathmatical geniuses? I can only use theories, I cannot make a specific mathmatical equation that would ever work properly...
- Kain Mercenary
- Luxer
- Posts: 201
- Joined: Mon May 23, 2005 4:21 pm
- Location: OMEGA HQ
The problem with your idea is that most people play to the system. Therefore, the scores don't show skill at the game, they show an ability to play within the system. In other words, if someone were to create a new scoring method that more accurately predicts who will win a game under the current system, that method will only maintain that accuracy if we continue to play under the current system. Adopting that new scoring method would, in all likelihood, change the behavior of players within games. It would also have a significant effect on the games players choose to play in as well as the opponents they would play against.GregM wrote:If we ignore the above (because there's really no way to compensate for it) how about this as an objective method of evaluating ranking systems: the ranking of players at the beginning of a game should predict the outcome of that game as often as possible.Dominator wrote:ALSO, someone pointed out that applying new equations to old data could be somewhat irrelevant. Players change the way they play or who they play with based on the ranking system in place. For example, when dustin's RAW system was in place everyone played more games because there was less risk in losing.
Another example is if you re-calculated last years NFL season. Imagine touchdowns worth 8 and field goals worth 4. The decisions that the coaches made during the season would not matter now.
Perhaps the following measurement would be good: if A is ranked above B, A should place higher than B in a game. The best ranking system is the one with the most successes in making this kind of prediction.
Random rankings give, of course, 50% accuracy. I ran a little test on games #500000-600000 and dustin's raw system seems to have a success rate of 59% -- that is, someone with higher raw will do better than someone with lower raw 59% of the time. If you look at seeds instead of raw (using raw to compare unseeded players) the success rate goes up to 61%. Not bad, given the unpredictability of Lux. But surely it can be improved.
Basically, this boils down to what Dominator said (someone said). Applying new formulas to old game data is dangerous. The only way to accurately measure the results of a scoring system is to put it into play and and make a judgement based on the effects.
- Kain Mercenary
- Luxer
- Posts: 201
- Joined: Mon May 23, 2005 4:21 pm
- Location: OMEGA HQ
My point was that applying a new formula to old data only predicts how someone would play under the old system and not necessarily the new one. Therefore, your percentage system is somewhat irrelevant in predicting behavior under a new system.GregM wrote:Agreed, but it's the best we've got short of a live test, which is expensive in time, and using past data should at least give some sense of what to expect.Kain Mercenary wrote:Applying new formulas to old game data is dangerous.
Perhaps running concurrent, officially supported (RAW: 1200, Bertrand: 42), systems is the best way to figure out what works and what doesn't. If enough people like an alternate system and its scoring algorithm, it may lead to that being the 'official' system.
I think the use of a percentage meter is an ineffective way of showing how well a system will perform when in actual use.
EDIT: Woah, and I'm currently in first under Bertrand's system! Perhaps it doesn't work so well.
However, I left out this condition, which drops predictive power to 59%:
----Bertrand wrote:But if the winner is already a "positive" player, and the average of the other players is negative, then it was an easy win, and I simply eliminate the base score from the calculation and only keep the skills score.
Clearly, but I can't think of a better way to get an initial estimate of a system's effectiveness. Any ideas?Kain Mercenary wrote:I think the use of a percentage meter is an ineffective way of showing how well a system will perform when in actual use.
OK, an example of how this analysis is flawed: eliminating skill points from Bertrand's system still gives a predictive accuracy of 61%, but it's very clear that this system could be exploited by cooperating to play a lot of games and having one person win, since their winnings don't diminish as their opponents get worse.
That's a brilliant insight. It's a very interesting way of comparing different scoring systems. So I couldn't resist trying it with my system.GregM wrote:If we ignore the above (because there's really no way to compensate for it) how about this as an objective method of evaluating ranking systems: the ranking of players at the beginning of a game should predict the outcome of that game as often as possible.
Perhaps the following measurement would be good: if A is ranked above B, A should place higher than B in a game. The best ranking system is the one with the most successes in making this kind of prediction.
Random rankings give, of course, 50% accuracy. I ran a little test on games #500000-600000 and dustin's raw system seems to have a success rate of 59% -- that is, someone with higher raw will do better than someone with lower raw 59% of the time. If you look at seeds instead of raw (using raw to compare unseeded players) the success rate goes up to 61%. Not bad, given the unpredictability of Lux. But surely it can be improved.
Did you filter out the matches where RAW was close to being equal? Since those matches can not be predicted, they represent "noise" that has to be removed from the final result.
My system does pretty well: using the long-term statistics (3 weeks worth of games), and filtering out the matches where the score difference was less than 20, the successful prediction percentage was 74%. This is significantly higher than chance, so it proves that my system actually means something.
- Kain Mercenary
- Luxer
- Posts: 201
- Joined: Mon May 23, 2005 4:21 pm
- Location: OMEGA HQ
How much of a spread is 20 when compared to the range (your high and low scores)? If it's to wide, you would end up testing only extremes.Bertrand wrote:That's a brilliant insight. It's a very interesting way of comparing different scoring systems. So I couldn't resist trying it with my system.GregM wrote:If we ignore the above (because there's really no way to compensate for it) how about this as an objective method of evaluating ranking systems: the ranking of players at the beginning of a game should predict the outcome of that game as often as possible.
Perhaps the following measurement would be good: if A is ranked above B, A should place higher than B in a game. The best ranking system is the one with the most successes in making this kind of prediction.
Random rankings give, of course, 50% accuracy. I ran a little test on games #500000-600000 and dustin's raw system seems to have a success rate of 59% -- that is, someone with higher raw will do better than someone with lower raw 59% of the time. If you look at seeds instead of raw (using raw to compare unseeded players) the success rate goes up to 61%. Not bad, given the unpredictability of Lux. But surely it can be improved.
Did you filter out the matches where RAW was close to being equal? Since those matches can not be predicted, they represent "noise" that has to be removed from the final result.
My system does pretty well: using the long-term statistics (3 weeks worth of games), and filtering out the matches where the score difference was less than 20, the successful prediction percentage was 74%. This is significantly higher than chance, so it proves that my system actually means something.
A spread of 20 represents 2 average wins in my system. It's an arbitrary value that intuitively looks good. In the current RAW system, I guess the equivalent would be something like 200 or 300 RAW points.Kain Mercenary wrote: How much of a spread is 20 when compared to the range (your high and low scores)? If it's to wide, you would end up testing only extremes.
Good point; a bigger ranking difference represents a stronger belief that A is better than B and so the evaluation system should take that into account. It would be interesting to plot the relationship between ranking difference and win probability for ranking systems under investigation.Bertrand wrote:Did you filter out the matches where RAW was close to being equal? Since those matches can not be predicted, they represent "noise" that has to be removed from the final result.
Edit:
Here's some output from my program, testing your system: probability of successful prediction of the outcome of a pairing versus absolute value of rating difference. As hoped, predictions get much more accurate for larger.
Code: Select all
(1, 4) - 8576 / 16435 = 52.1%
(4, 9) - 10777 / 19681 = 54.7%
(9, 16) - 10982 / 18448 = 59.5%
(16, 25) - 8443 / 12551 = 67.2%
(25, 36) - 6229 / 8381 = 74.3%
(36, 49) - 3787 / 4648 = 81.4%
(49, 64) - 1335 / 1528 = 87.3%
(64, 81) - 286 / 315 = 90.7%
(81, 100) - 42 / 45 = 93.3%
Code: Select all
(0, 50) - 10297 / 19410 = 53.0%
(50, 100) - 8518 / 14841 = 57.3%
(100, 150) - 6723 / 11216 = 59.9%
(150, 200) - 5339 / 8657 = 61.6%
(200, 300) - 7502 / 11624 = 64.5%
(300, 400) - 4392 / 6481 = 67.7%
(400, 500) - 2243 / 3161 = 70.9%
(500, 700) - 1474 / 1876 = 78.5%
(700, 1000) - 204 / 247 = 82.5%
(1000, 1500) - 6 / 7 = 85.7%
how about bots? are they included?GregM wrote:Good point; a bigger ranking difference represents a stronger belief that A is better than B and so the evaluation system should take that into account. It would be interesting to plot the relationship between ranking difference and win probability for ranking systems under investigation.Bertrand wrote:Did you filter out the matches where RAW was close to being equal? Since those matches can not be predicted, they represent "noise" that has to be removed from the final result.
Who is online
Users browsing this forum: No registered users and 122 guests