f_alk
![](./modules/PNphpBB2/images/avatars/upload/4373953884d3fe6a65f5e9.gif)
Joined: Sep 30, 2005
|
  Posted:
Jul 27, 2011 - 00:52 |
|
so, we use a different approach to our statistics. I can live with that, especially since your CDorf example shows we are quite close. But even then, your approach has even more teams who show a significant difference between their distribution of opposing team races comparing with and without mirror matches.
So, your approach shows more impact of the discouraging of mirror matches.
I use "chaos play the most matches and the most mirror matches" only as a proof for the well known fact that mirror matches are slightly discouraged, and you can see that little factor in there. You have less Chaos mirror matches than you would expect to happen , calculated from the amount of chaos-involving matches and absolute matches.
Towards your argument of taking into account the team numbers of each race: yes, I thought of that, but unfortunately we have no idea about the spread of games/team value you gave the averages for. Of course, we know that your example couldbe detected - as we have the numbers of teams for the "no mirror matches" (nmm) and the "with mirror matches" (wmm) for the races, and there is little change in there. So for that we would see in the "best case":
chaos teams -- wmm: 1 -- nmm:1
DE teams -- wmm 150: -- nmm: 50
So, a big change in the wmm and nmm number would indicate something to look into, but we don't have that here. |
|
|
jimjo
Joined: Apr 21, 2005
|
  Posted:
Jul 27, 2011 - 00:59 |
|
when i look at the causalities per game
i expect that when all 'Home cas and all 'Away cas
counted together the result should be very close (or equal) but it isn't.
or didn't i get it?
----------------- |
|
|
Sutherlands
Joined: Aug 01, 2009
|
  Posted:
Jul 27, 2011 - 01:15 |
|
@jimjo: I got the same result for both columns. You're using the numbers and not per-game numbers, right?
@f_alk: It's not a matter of detecting it. It's simply a variable that you haven't accounted for in your model, and thus your model is incomplete. It's also a very important variable. Simply put, there's no way to tell what effect the (very small) mirror match modifier has simply by looking at the output of the scheduler. We have to know what options for scheduling it had to work with.
"Statistics don't lie, but liars make statistics." |
|
|
The_Murker
![](./modules/PNphpBB2/images/avatars/upload/20451654904e064e042d2bd.gif)
Joined: Jan 30, 2011
|
  Posted:
Jul 27, 2011 - 01:16 |
|
Thanks very much for those stats. The ability for people to mull things over and ask questions is much appreciated. |
_________________
Join the wait-list. Watch the action. Leave the Empire. Come to Bretonnia! |
|
f_alk
![](./modules/PNphpBB2/images/avatars/upload/4373953884d3fe6a65f5e9.gif)
Joined: Sep 30, 2005
|
  Posted:
Jul 27, 2011 - 01:44 |
|
Sutherlands wrote: | @f_alk: It's not a matter of detecting it. It's simply a variable that you haven't accounted for in your model, and thus your model is incomplete. It's also a very important variable. Simply put, there's no way to tell what effect the (very small) mirror match modifier has simply by looking at the output of the scheduler. We have to know what options for scheduling it had to work with.
|
True, the model is incomplete. But that's a strawman argument. If I can show that this additional variable has little or no effect, then I can neglect it (or widen the error bars of my result accordingly).
I will give more tomorrow, it is too late here at the moment. I would love to see your analysis though, and your proof that the variable is very important, from the numbers we have. |
|
|
f_alk
![](./modules/PNphpBB2/images/avatars/upload/4373953884d3fe6a65f5e9.gif)
Joined: Sep 30, 2005
|
  Posted:
Jul 27, 2011 - 01:58 |
|
Just one quick thing:
If you take your "chance to meet" from the last page, and add those up over all races... you get 193%. Sure mine is not much better, just shows we need a better measure first place. Any ideas welcome
Edit:
Instead dividing by 1785 we should divide of course by 1785-118/2 ... 118 being the number of mirror matches. then it sums up nicely to 200%, so we can divide it by 2 to get a 100% chance.
Edit2:
comparing these numbers "mirror match numbers" to the "No mirror match numbers" ... you can create the ratio of difference to "mirror match numbers", and you will see that all those teams without mirror matches come up with the same negative value for that ratio (about -3.5%) ... now it is really getting late, I'll leave you with this. |
|
|
Sutherlands
Joined: Aug 01, 2009
|
  Posted:
Jul 27, 2011 - 19:34 |
|
1) You haven't shown that the variable has little or no effect. The example I gave shows that it is important. You say that you can detect it in that case... well yes, but that's unimportant to the model. The number of teams of a specific race that are playing games has a very real effect on how many mirror matches there are. The lower the number of teams that are playing the games, the lower the chance of a mirror match is.
2) Yes, you get 193% if you add up all the races. First, that's not "chance to meet", it's "chance to see". And given that a mirror match you only see 1 race, it makes sense for it to sum to less than 200%. Pointing out that it doesn't is a red herring, and has no bearing on the statistics quoted. |
|
|
stej
Joined: Jan 05, 2009
|
  Posted:
Jul 28, 2011 - 00:53 |
|
Chi squared goodness of fit test please.... |
|
|
Aries
Joined: Apr 01, 2004
|
  Posted:
Jul 28, 2011 - 01:00 |
|
your statistics about mirrors are completely hypothetical without knowing which teams/tvs activated every single time, and what options the bot had before choosing.
if only 2 2000tv teams activate and both chaos,and 2 tv1000 always mirror. if 1 chaos 2ktv and 1 other 2ktv activate, always non mirror. so seeing what was played, doesnt tell you what was activated. therefore, youre making a large presumption based on numbers that dont show everything that happened. |
|
|
f_alk
![](./modules/PNphpBB2/images/avatars/upload/4373953884d3fe6a65f5e9.gif)
Joined: Sep 30, 2005
|
  Posted:
Jul 28, 2011 - 14:32 |
|
Sutherlands wrote: | 1) You haven't shown that the variable has little or no effect. The example I gave shows that it is important. You say that you can detect it in that case... well yes, but that's unimportant to the model. The number of teams of a specific race that are playing games has a very real effect on how many mirror matches there are. The lower the number of teams that are playing the games, the lower the chance of a mirror match is. |
I have shown that the variable is very small. Also, the effect is best/worst quadratic, as it is a matching effect. In that the effect of a small variable will stay small.
Your example was an extreme example, and I could show that we are not in that extreme regime. Yes, there is an effect, we agree on that. And it seems we agree that we can actually check it ourselves: How many teams have played mirror matches only (the difference of the teams counted), and of how many mirror matches have been played? You can create the "most extreme" example that complies with that set of data, and it won't be that extreme.
Quote: | 2) Yes, you get 193% if you add up all the races. First, that's not "chance to meet", it's "chance to see". And given that a mirror match you only see 1 race, it makes sense for it to sum to less than 200%. Pointing out that it doesn't is a red herring, and has no bearing on the statistics quoted. |
No, whenever I add up "chances" over all possibilities, I must end with 100%.
Else I can end up with something impossible, which - as the name says - is impossible.
I don't have much time this and the next week, I will see if I can get more out of that. |
|
|
Sutherlands
Joined: Aug 01, 2009
|
  Posted:
Jul 28, 2011 - 19:34 |
|
f_alk wrote: |
I have shown that the variable is very small. | You have? I see no such proof.
f_alk wrote: | Also, the effect is best/worst quadratic, | Wha??
f_alk wrote: | No, whenever I add up "chances" over all possibilities, I must end with 100%. | Seems kind of odd, then, that you modified it to add up to 200%.
f_alk wrote: | Else I can end up with something impossible, which - as the name says - is impossible. | No, you end up with percentages that were not made to sum together. If I say that the number of days that saw bright sun during a year were 300, and the days that saw rain were 150, that's obviously more than the number of days in a year, and that's accounted for by the fact that some can have both (and same can have neither).
The point of the games/team is that you're looking at the completely wrong set of stats and comparing them as if they're comparable, but they're not. I'm telling you this and you're just putting your fingers in your ears because it doesn't go with what you want. You wouldn't expect the percentage of mirror matches to be played to be the same percentage as the number of games played, you would expect it to be a percentage related to the different numbers of teams. All you've said is "we're not that extreme example, so it's ignorable." No, no it's not. And once again, you can't make any conclusions from this data, as you need the input data to the scheduler, which is not here. |
|
|
licker
![](./modules/PNphpBB2/images/avatars/upload/166509707555ef89b25b521.jpg)
Joined: Jul 10, 2009
|
  Posted:
Jul 28, 2011 - 20:00 |
|
Quote: | And once again, you can't make any conclusions from this data,... |
Of course you can. What is it with this notion that you can't make a conclusion without a complete data set? Utter bollocks.
Now f_alk may or may not be making a supportable conclusion, that's an entirely different question though, and one I don't care to debate (again) because it's clear from the statistics, and clear from past input about the scheduler that mirror matches do not, and are not intended to happen at the frequency one would expect in an unbiased scheduler. (and over enough data points you can start to stop worrying about the stupidity of 'did only 2 TV 2000 teams show up', so stop with that strawman)
Which is just saying, the scheduler *has* a bias in it. An intentional bias, it's doing what it's coded to do. Kinda pointless really, but the question has never been is the scheduler doing what it's coded to do, the question has always been, what should the scheduler do?
Anyway, continue, until an admin finds another hair up their butt and closes this thread too. |
|
|
Aries
Joined: Apr 01, 2004
|
  Posted:
Jul 28, 2011 - 20:16 |
|
Until the data that shows timestamped all teams that activated every single time you can at best very roughly and inaccurately about mirrors. Assumptions would be based on incomplete data. If I have 10 teams I activate each time only 1 gets a game but the scheduler sees 9 other teams that the data doesn't show here. Compound that by everyone that played box and you've got an uncontrolled invisible variable that makes any debate absolutely ludicrous.
Get Christer or Someone involved in the coding to address your concerns with scheduling ... These stats have NOTHING to do with scheduling and only show exactly what they show. |
|
|
The_Murker
![](./modules/PNphpBB2/images/avatars/upload/20451654904e064e042d2bd.gif)
Joined: Jan 30, 2011
|
  Posted:
Jul 28, 2011 - 20:18 |
|
I would agree with the belief that any conclusions drawn from the stats aren't very valid without knowing the options given to the scheduler. We can therefore speculate a bunch on what we see here, and not really prove each other wrong, certainly not without being more systematic. Someone needs a proper hypothesis for parties to offer data to support or disprove. Here's an example of one, but you guys could do better. I'd suggest one hypothesis per thread, to avoid confusion and not get sidetracked.
Some facts from the data:
Fact: Chaos and Chaos Dwarfs play more games than any of the other races, and dominate the high TV range.
Resulting Assumption: Therefore, the scheduler will be FORCED to pair more mirror matches between those two races, as it has fewer choices, most notably in the high TV ranges. (IMO, this is why there appears to be "enough" mirrors for the those two races)
Hypothesis: If the scheduler were racially blind, the number of mirror matches played by Chaos and Chaos Dwarf teams would increase SIGNIFICANTLY. (the word "significant" would have to be quantified. I submit a 15% increase in mirrors played by BASH teams would change their exposure to injury in a "team affecting" way) |
|
|
Sutherlands
Joined: Aug 01, 2009
|
  Posted:
Jul 28, 2011 - 21:15 |
|
NO!
Fact: Likelyhood of mirror matches is closer to the number of Teams than it is to the number of games played. |
|
|
|
| |