Tournament World Cup of Randbats 2023 - Qualifying Round

Status
Not open for further replies.
Wigglytuff said:
how are these calculated? 0.714 seems like 71.4/100 but 0.287 seems to come out of nowhere other than possibly being 1 - 0.714 with a rounding error. is this something similar to strength of schedule / what are its advantages over strength of schedule?
Click to expand...
the exact formula is: [p(1)*1-p(2)] / [p(1)*1-p(2)] + [p(2)*1-p(1)] - IE, the chance that Player 1 wins and player 2 loses, divided by the combined chance of both that and the inverse. Because Trade had exactly a 50% winrate, it cancelled itself out in the final result.
For a better example, a hypothetical game between GXE (64.7%) and BigWes95 (68.8%) would yield a 0.455 - 0.545 split
I'll take a look at SoS and see if I could incorporate it

Wigglytuff said:
if "WR" doesn't take into account completed games and restarts from 0, what are the above fractional points that are then used to calculate "WR" derived from?
Click to expand...
The first 2 columns award full points (1-0) to the games that already have been completed. This is why no team has less points than their current score.
The 2nd set use the exact same seed probabilities, but try to "predict" the results of games that we already know who won. Thats why US South has only 36 points, even though they have 38 wins

Wigglytuff said:
i agree that it'd be difficult to find gxe's of players, but how did you end up with sub 40 gxe's for any team? it takes a concerted effort of forfeiting every game to hit below 50, let alone the 11 mexico has in your stats
Click to expand...
That 11 represents 11 wins in this tournament, based on GxE values of each player in team mexico. There is only 1 player in the entire tournament with a below 50% GxE, and tbh I was tempted to just hard set them to 50% like I did with the many players with no ladder stats
 
Dark Pulse said:
"G%" isn't GxE, its game win percentage for the team across the tournament, basically its just G.win / G.win+G.lose
Only 1 player in the tournament has a GxE below 50%.
Click to expand...
1692416218378.png

this column on the predictions end titled GxE is the one i was referring to

Dark Pulse said:
[p(1)*1-p(2)] / [p(1)*1-p(2)] + [p(2)*1-p(1)]
Click to expand...
did you mean
1692416823783.png
?

because the formula as written would give 1 + p(2)-p(1)

i'm still unsure of what predictive advantage this would have over strength of schedule, because the strength of the opponents faced is completely unaccounted for (something that is very relevant in a scenario like this where opponents don't necessarily share the opponents of their opponents)

Dark Pulse said:
The 2nd set use the exact same seed probabilities, but try to "predict" the results of games that we already know who won. Thats why US South has only 36 points, even though they have 38 wins
Click to expand...
if the seed probabilities are the same as calculated from actual results and the formula applied is the same, what accounts for the difference in the prediction vs actual results?

sorry if this sounds like an interrogation. i'm all for the power of statistics in helping players/managers evaluate player strength in high variance formats like rands, but the ones here seem to be lacking compared to standard predictive models
 
Wigglytuff said:
View attachment 544273
this column on the predictions end titled GxE is the one i was referring to
Click to expand...
Yep, realised after posting. Original post edited with explanation

Wigglytuff said:
did you mean
View attachment 544275?

because the formula as written would give 1 + p(2)-p(1)
Click to expand...
1692417756852.png

referenced cells are either the GxE or the winrate of players 1 & 2, depending on which seed value I'm using. As far as I can tell, 1+P2-P1 gives a completely different result.

Wigglytuff said:
i'm still unsure of what predictive advantage this would have over strength of schedule, because the strength of the opponents faced is completely unaccounted for (something that is very relevant in a scenario like this where opponents don't necessarily share the opponents of their opponents)
Click to expand...
There probably isn't any. I'm not a math major and have no formal education in statistics, only what I've picked up while learning if for other high-variance games I play. Most of this is pure intuition written up on the fly (why I added the caveat of "simple" to my initial post) - If someone wants to either take what I've got and improve it, or collaborate/suggest improvements I could make, I'd be eager to take the conversation to DMs and discuss/send my file across

Wigglytuff said:
if the seed probabilities are the same as calculated from actual results and the formula applied is the same, what accounts for the difference in the prediction vs actual results?
Click to expand...
Not quite sure I follow here. Do you mean the difference between the pair of "predictions" columns vs the pair of "recreated" columns? Or do you mean the difference between S.win and all 4 predictions
Wigglytuff said:
sorry if this sounds like an interrogation. i'm all for the power of statistics in helping players/managers evaluate player strength in high variance formats like rands, but the ones here seem to be lacking compared to standard predictive models
Click to expand...
You're totally fine, and as said they almost certainly are. Would love to go through this in more detail sometime if you're willing to spend time on it
 
regrettably posting again about calling act on Team Venezuela:

- took a full week after 2 pings and a prior act post to provide a sub
- said sub did not respond to my scheduling before almost a full day had passed and then offered a time quite late for me (which also had serious constraints as I mentioned prior as scheduling late left me with less availability)
- offered another alternative time the following day that was never responded to

Given the circumstances I think this falls under clear guidelines for an act call.

cc hosts: Syrinix sharpclaw
 
 
 
Status
Not open for further replies.
Back
Top