Predictor Week 6

[LegendaryGT]

The predictor was a perfect 6-0 ATS in last week's thread, and went 4-1 on the money line (only missing the Duke game, and not picking a winner in the UF game). It also probably has more memories of Saturday than I do.

GT vs CLEM
Week 6 of 2015

GT - Offense
Yardage: 401 ypg
Performance: 13 (higher is better)
Scoring: 34
Efficiency: 11 (lower is better)
Performance: 36 (higher is better)
GT - Defense
Yardage: 335 ypg allowed
Performance: 30 (higher is better)
Scoring: 28
Efficiency: 11 (higher is better)
Performance: -18 (higher is better)

CLEM - Offense
Yardage: 361 ypg
Performance: 11 (higher is better)
Scoring: 28
Efficiency: 12 (lower is better)
Performance: 25 (higher is better)
CLEM - Defense
Yardage: 338 ypg allowed
Performance: 31 (higher is better)
Scoring: 16
Efficiency: 20 (higher is better)
Performance: 64 (higher is better)

Prediction:

Pure statistical: GT 22 (300 yards), CLEM 30 (304 yards)
GT best case: GT 27 (387 yards), CLEM 15 (253 yards)
CLEM best case: GT 15 (239 yards), CLEM 41 (380 yards)

GT talent: 5051
CLEM talent: 10803
GT talent of schedule: 6406
CLEM talent of schedule: 6035

Suggested talent adjustment: 53% towards CLEM best case
Final: GT 18 (267 yards), CLEM 36 (345 yards)

And the rest of the schedule:

GT 17, PITT 21
GT 15, FSU 28
GT 60, UVA 26
GT 54, VT 31
GT 42, MIA 30
GT 37, UGA 45

It flipped its forecast in the Pitt game, so it thinks we're going 5-7. Bowl waver inbound.

#1: ('FLA', [0.9875, 117, 3, 5, 0])
#2: ('OKLA', [0.975, 114, 6, 4, 0])
#3: ('FSU', [0.9667, 112, 8, 4, 0])
#4: ('IOWA', [0.9583, 110, 10, 4, 0])
#5: ('OSU', [0.9542, 109, 11, 5, 0])
#6: ('NW', [0.9417, 106, 14, 4, 0])
#7: ('BAY', [0.9417, 106, 14, 3, 0])
#8: ('CLEM', [0.9417, 106, 14, 2, 0])
#9: ('LSU', [0.9333, 104, 16, 4, 0])
#10: ('UTAH', [0.9333, 104, 16, 4, 0])
#11: ('TAMU', [0.9292, 103, 17, 3, 0])
#12: ('ALA', [0.9, 120, 0, 4, 1])
#13: ('MICH', [0.8958, 119, 1, 4, 1])
#14: ('STAN', [0.8833, 116, 4, 4, 1])
#15: ('TCU', [0.875, 90, 30, 4, 0])
#16: ('USC', [0.8667, 118, 2, 3, 1])
#17: ('CAL', [0.8542, 85, 35, 4, 0])
#18: ('HOU', [0.8542, 85, 35, 3, 0])
#19: ('UGA', [0.8333, 110, 10, 3, 1])
#20: ('TOL', [0.8292, 79, 41, 4, 0])
#21: ('NAVY', [0.8292, 79, 41, 3, 0])
#22: ('ND', [0.825, 102, 18, 4, 1])
#23: ('MISS', [0.8042, 103, 17, 3, 1])
#24: ('MSU', [0.7833, 68, 52, 5, 0])
#25: ('TEM', [0.7708, 65, 55, 3, 0])

#59: ('GT', [0.5042, 91, 29, 1, 3])

Made an update to the predictor poll to make it more human readable and to reduce the weird outliers. The format is now, from left to right, composite rating, simulated wins, simulated losses, actual FBS wins*, actual FBS losses*.

* The criteria for being counted as an FBS school for the purposes of the algorithm actually excludes Georgia State and Charlotte, so totals may not match for teams on their schedules.

 

[Akinji07]

Could you do uga Tennessee?

 

[LegendaryGT]

Could you do uga Tennessee?

Tennessee is one of those weird outliers. They are clearly and obviously bad in some important ways, but the predictor loves them. The difference between their simulated win percentage and their actual win percentage is 65.3%, which is the highest in FBS by a very very wide margin. That means this prediction is very likely to be way bad.

UGA vs TENN
Week 6 of 2015

UGA - Offense
Yardage: 421 ypg
Performance: 35 (higher is better)
Scoring: 36
Efficiency: 11 (lower is better)
Performance: 27 (higher is better)
UGA - Defense
Yardage: 326 ypg allowed
Performance: 19 (higher is better)
Scoring: 21
Efficiency: 15 (higher is better)
Performance: -9 (higher is better)

TENN - Offense
Yardage: 404 ypg
Performance: 9 (higher is better)
Scoring: 30
Efficiency: 13 (lower is better)
Performance: 22 (higher is better)
TENN - Defense
Yardage: 436 ypg allowed
Performance: 15 (higher is better)
Scoring: 24
Efficiency: 17 (higher is better)
Performance: 29 (higher is better)

Prediction:

Pure statistical: UGA 32 (492 yards), TENN 46 (341 yards)
UGA best case: UGA 46 (591 yards), TENN 32 (291 yards)
TENN best case: UGA 22 (352 yards), TENN 61 (383 yards)

UGA talent: 11371
TENN talent: 10473
UGA talent of schedule: 7968
TENN talent of schedule: 7795

Suggested talent adjustment: 10% towards UGA best case
Final: UGA 33 (503 yards), TENN 45 (336 yards)

 

[GTLiebs]

I love the double digit score difference toward Tenn while having them get almost 200 less yards of offense. Guess that is part of the shenanigans you mentioned.

 

[coit]

The predictor was a perfect 6-0 ATS in last week's thread, and went 4-1 on the money line

Wow that's not too shabby!

 

[LegendaryGT]

I love the double digit score difference toward Tenn while having them get almost 200 less yards of offense. Guess that is part of the shenanigans you mentioned.

It's an efficiency calculation that's tweaked out because of UGA crapping their pants against Alabama. 4 turnovers by the QB including a pick six, on top of a blocked punt for a TD. The predictor is convinced UGA is turnover prone and opponents are going to get tons of free points.

 

[GTCrew4b]

öööö this predictor.

Predict this: *farts at predictor*

 

[GT1992]

The Predictor is predicting no bowl now? Needs tweaking now.

 

[smokey_wasp]

Bud Foster is gonna keel over if we put 54 on VT.

 

[Akinji07]

I wonder if you can build an algorithm to make offensive and defensive calls based on history of opponent's playcalling and have it play against teams on our schedule with the current squad then we can replace it with Paul Johnson if it starts performing better.

 

[coit]

I wonder if you can build an algorithm to make offensive and defensive calls based on history of opponent's playcalling and have it play against teams on our schedule with the current squad then we can replace it with Paul Johnson if it starts performing better.

Somewhat related, as a you're at GT, you can do that sort of thing, shouldn't we be using game theory on defense to predict what the other team is going to do and defend against it?

We could have a crack team of game theory experts, right?

 

[Akinji07]

Somewhat related, as a you're at GT, you can do that sort of thing, shouldn't we be using game theory on defense to predict what the other team is going to do and defend against it?

We could have a crack team of game theory experts, right?

Even Paul Johnson thinks that's too öööööööy.

 

[Akinji07]

Wait a minute... ööööööö is a bad word???

 

[coit]

G immicky?

 

[Akinji07]

.

Yeah.

 

[LegendaryGT]

If there's one thing I've learned after a half decade of this, it's that college football data is almost entirely noise. I'm not sure you would ever be able to generate enough of it at a high enough quality to automate playcalling at the same level as a decent coach. I'm confident that the very best coaches will always outperform their mechanical competitors. It's not really a question of whether the computer can out-theory you, it's just a matter of whether it can gather better data. Maybe in the superfuture when we have the ball and all of the players tracked at the milimeter/milisecond level we can put all the data into a PS15 and play it through Madden 2065 to test this theory.

 

[ScionOfSouthland]

Preidctor for VT might be a little hard because we see a different defense than every other team they see. The sample for that one may be better adjusted by using their performance against us and Ohio State based on our skill level in that given year.

 

[LegendaryGT]

GT vs VT
Week 6 of 2015

GT - Offense
Yardage: 401 ypg
Performance: 13 (higher is better)
Scoring: 34
Efficiency: 11 (lower is better)
Performance: 36 (higher is better)
GT - Defense
Yardage: 335 ypg allowed
Performance: 30 (higher is better)
Scoring: 28
Efficiency: 11 (higher is better)
Performance: -18 (higher is better)

VT - Offense
Yardage: 337 ypg
Performance: -2 (higher is better)
Scoring: 29
Efficiency: 11 (lower is better)
Performance: 20 (higher is better)
VT - Defense
Yardage: 378 ypg allowed
Performance: 7 (higher is better)
Scoring: 29
Efficiency: 12 (higher is better)
Performance: -7 (higher is better)

Prediction:

Pure statistical: GT 61 (408 yards), VT 30 (254 yards)
GT best case: GT 79 (430 yards), VT 24 (226 yards)
VT best case: GT 35 (372 yards), VT 35 (327 yards)

GT talent: 5051
VT talent: 7716
GT talent of schedule: 6406
VT talent of schedule: 6328

Suggested talent adjustment: 25% towards VT best case
Final: GT 54 (398 yards), VT 31 (273 yards)

Full prediction for reference. VT's overall performance in 2014:

VT - Offense
Yardage: 348 ypg
Performance: -9 (higher is better)
Scoring: 22
Efficiency: 15 (lower is better)
Performance: -7 (higher is better)
VT - Defense
Yardage: 333 ypg allowed
Performance: 25 (higher is better)
Scoring: 20
Efficiency: 15 (higher is better)
Performance: 18 (higher is better)

First 6 games of 2014:

VT - Offense
Yardage: 389 ypg
Performance: -7 (higher is better)
Scoring: 29
Efficiency: 13 (lower is better)
Performance: 3 (higher is better)
VT - Defense
Yardage: 335 ypg allowed
Performance: 48 (higher is better)
Scoring: 22
Efficiency: 15 (higher is better)
Performance: 24 (higher is better)

Even though VT often presents us with different fronts, it's clear, statistically at least, that this year's version is not nearly as good as previous editions. 50+ would surprise me, for sure, but I expect that dropoff to translate even to us.

 

[GruffyMcGuiness]

Can you do UCLA - Stanford? Gotta prepare for yet another loss to the trees.

 

[LegendaryGT]

UCLA vs STAN
Week 6 of 2014

UCLA - Offense
Yardage: 432 ypg
Performance: 8 (higher is better)
Scoring: 34
Efficiency: 12 (lower is better)
Performance: 3 (higher is better)
UCLA - Defense
Yardage: 386 ypg allowed
Performance: -2 (higher is better)
Scoring: 22
Efficiency: 17 (higher is better)
Performance: 7 (higher is better)

STAN - Offense
Yardage: 454 ypg
Performance: 28 (higher is better)
Scoring: 35
Efficiency: 12 (lower is better)
Performance: 20 (higher is better)
STAN - Defense
Yardage: 323 ypg allowed
Performance: 25 (higher is better)
Scoring: 19
Efficiency: 17 (higher is better)
Performance: 29 (higher is better)

Prediction:

Pure statistical: UCLA 17 (324 yards), STAN 35 (515 yards)
UCLA best case: UCLA 21 (358 yards), STAN 33 (463 yards)
STAN best case: UCLA 12 (268 yards), STAN 40 (594 yards)

UCLA talent: 9043
STAN talent: 8965
UCLA talent of schedule: 5128
STAN talent of schedule: 6303

Suggested talent adjustment: 10% towards STAN best case
Final: UCLA 16 (318 yards), STAN 36 (523 yards)

#14: ('STAN', [0.8833, 116, 4, 4, 1])

#35: ('UCLA', [0.725, 78, 42, 4, 1])

Stanford's simulated winning percentage is 16.6% higher than its actual winning percentage. UCLA's simulated winning percentage is 15% lower than its actual winning percentage. I'm still trying to decide whether that's meaningful in any way, but it is interesting enough.