Return of the algorithm

[Hobbes]

I've revived my tipping algorithm after a few years' absence.

The gist is that each team has a rating which reflects their expected result against an average team on neutral ground. All past matches are scored against teams' current rating, so matches can be retrospectively re-evaluated. All old results deteriorate at 10% per round. I figure that halfway through the season is enough time to consider last year's results irrelevant.

Home ground advantage is calculated as: none for same-state teams, 12 points if a team from SA or WA is involved, 6 points if a team from NSW or Qld is involved. The highest modifier is taken, but they don't add.

I also condense margins greater than 50 points - half of the difference between the margin and 50 is subtracted.

Current ratings:

Brisbane 25.84
Collingwood 20.77
Melbourne 17.25
Western Bulldogs 14.47
Adelaide 12.51
Port Adelaide 11.22
Fremantle 5.7
Essendon 5.45
Geelong 4.59
St Kilda 3.72
Gold Coast 0.91
Richmond -2.38
Sydney -3.43
Carlton -5.55
GWS -9.29
Hawthorn -17.51
North Melbourne -30.03
West Coast -54.91

Current projected ladder (just measured in wins)

Collingwood 18.1
Brisbane 16.83
Port Adelaide 16.12
Western Bulldogs 15.39
Melbourne 15.29
St Kilda 13.61
Adelaide 13.54
Essendon 13.28
Fremantle 12.76
Gold Coast 10.84
Geelong 10.59
Sydney 10.16
Carlton 9.84
Richmond 9.14
GWS 8.69
Hawthorn 6.32
North Melbourne 4.5
West Coast 1.97

(West Coast is tipped to be as good as this because my algorithm won't let any team be less than a 5% chance. This may be an error in this case.)

And tips for round 12:

Melbourne +23 v Carlton
Port Adelaide +41 v Hawthorn
West Coast v Collingwood +64
Western Bulldogs +10 v Geelong
Gold Coast +1 v Adelaide
GWS v Richmond +1
Essendon +36 v North Melbourne

Gold Coast-Adelaide and GWS-Richmond are painfully tough to tip.

 

[SBYM]

Noice.

Melbourne will pump the Bombers in the EF for sure.

 

[Hobbes]

Noice.

Melbourne will pump the Bombers in the EF for sure.

On the current ratings and projected ladder, we get:

Collingwood d Bulldogs
Brisbane d Port
St Kilda d Adelaide
Melbourne d Essendon

Bulldogs d Melbourne
Port d St Kilda

Collingwood d Port
Brisbane d Melbourne

Brisbane d Collingwood (I don't assess any home ground advantage for the GF)

 

[BigPond101]

Your algorithm needs a human touch to know after a brief honeymoon period under Ross Lyon St Kilda are in the midst of an almighty collapse will be lucky to win 2 more games for tne year let alone make the 8

 

[Hobbes]

Your algorithm needs a human touch to know after a brief honeymoon period under Ross Lyon St Kilda are in the midst of an almighty collapse will be lucky to win 2 more games for tne year let alone make the 8

I think my algorithm responds more quickly to form shifts than many other tipping methods.

 

[maskmcgee]

All past matches are scored against teams' current rating, so matches can be retrospectively re-evaluated.

Good, so prove it works by retrospectively re-evaluating every game for the season so far so we can see how many tips it got right to see if its any good.

 

[Mop]

Heh heh heh. We are biblical levels of bad.


Nice method though

 

[Hobbes]

Heh heh heh. We are biblical levels of bad.


Nice method though

I can't remember if I did it retrospectively, but GWS in 2012 might have been of similar calibre. It's pretty dire. Most teams are 110% or more to win, or at least would be if I didn't cap at 95%.

 

[Mop]

I can't remember if I did it retrospectively, but GWS in 2012 might have been of similar calibre. It's pretty dire. Most teams are 110% or more to win, or at least would be if I didn't cap at 95%.

Heh heh.

I wonder who'd be worse - us now or Melbourne circa 2008/9 - the Jack Watts era

 

[Hobbes]

Good, so prove it works by retrospectively re-evaluating every game for the season so far so we can see how many tips it got right to see if its any good.

There's a few hours work, and having revived and modified the method I'd have to decide how last year's results impact. (After 11 rounds I feel fine ignoring last year.). I'll have a go though.

(In the dim, dark, pre-lockdown past I shared my spreadsheet with Final Siren. I don't know if he's tried keeping it running - he might have stats.) Or, if you dig deep enough on Bigfooty, you might find recorded results for 2015-2018 or something.

 

[Hobbes]

Heh heh.

I wonder who'd be worse - us now or Melbourne circa 2008/9 - the Jack Watts era

The worst teams in the AFL era are GWS 2012 and Fitzroy 1996 - both with a percentage under 50.

 

[Mop]

The worst teams in the AFL era are GWS 2012 and Fitzroy 1996 - both with a percentage under 50.

Ah okay fair fair.




I reckon we'd maybe JUST beat GWS 12.

 

[Fadge]

Good, so prove it works by retrospectively re-evaluating every game for the season so far so we can see how many tips it got right to see if its any good.

Of course if an algorithm uses recent results to retrospectively select who wins those recent games, it will be better than most if not all other approaches given it has the benefit of hindsight.

The test is to use the algorithm to select the winners for each upcoming round - maybe the OP can use this thread to share on a week by week basis?

 

[Hobbes]

Of course if an algorithm uses recent results to retrospectively select who wins those recent games, it will be better than most if not all other approaches given it has the benefit of hindsight.

The test is to use the algorithm to select the winners for each upcoming round - maybe the OP can use this thread to share on a week by week basis?

I'm going to run the algorithm from round 1, using the results from Round 16 last year, onward. I won't use future results for a retrospective prediction. I will post retrospective selections for earlier rounds, and I'm prepared to look embarrassed doing so, but what I have done in the past (and will do here) is post predictions every week, and keep score.

 

[Hobbes]

Retrospective tips - Round 1.

Initial ratings

Geelong 35.54 (I'm prepared to see this blow up in my face)
Sydney 12.15
Richmond 11.52
Port Adelaide 10.11
Melbourne 8.48
Collingwood 6.33
Fremantle 4.3
Brisbane 1.49
Carlton 0.95
Western Bulldogs -0.31
Gold Coast -2.28
St Kilda -3.38
Adelaide -4.37
Hawthorn -4.57
Essendon -4.98
GWS -8.82
West Coast -10.88
North Melbourne -15.6

I'm using (initially) rounds 16-onward from last year. Byes are entered as a zero round, and will affect scores (probably bringing them a little toward zero). An additional bye is counted after the grand final. Each round I'll remove one of last year's rounds from the data, starting with the oldest.

(all of these can be inferred by the method and the ratings. If the algorithm picks a draw, I'll round it out to a 1 point margin one way or the other.)

Richmond +11 v Carlton
Geelong +29 v Collingwood
North Melbourne +7 v West Coast
Port Adelaide +21 v Brisbane
Melbourne +13 v Western Bulldogs
Gold Coast v Sydney +8
GWS +8 v Adelaide
Hawthorn +1 v Essendon
St Kilda +7 v Fremantle

6.5/9 from round 1.

 

[maskmcgee]

What about last round, did it tip upsets that were in retrospect not so much upsets as teams moving in different directions? Such as Melbourne v Fremantle and St Kilda v. Hawthorn?

 

[Fadge]

Retrospective tips - Round 1.

Initial ratings

Geelong 35.54 (I'm prepared to see this blow up in my face)
Sydney 12.15
Richmond 11.52
Port Adelaide 10.11
Melbourne 8.48
Collingwood 6.33
Fremantle 4.3
Brisbane 1.49
Carlton 0.95
Western Bulldogs -0.31
Gold Coast -2.28
St Kilda -3.38
Adelaide -4.37
Hawthorn -4.57
Essendon -4.98
GWS -8.82
West Coast -10.88
North Melbourne -15.6

I'm using (initially) rounds 16-onward from last year. Byes are entered as a zero round, and will affect scores (probably bringing them a little toward zero). An additional bye is counted after the grand final. Each round I'll remove one of last year's rounds from the data, starting with the oldest.

(all of these can be inferred by the method and the ratings. If the algorithm picks a draw, I'll round it out to a 1 point margin one way or the other.)

Richmond +11 v Carlton
Geelong +29 v Collingwood
North Melbourne +7 v West Coast
Port Adelaide +21 v Brisbane
Melbourne +13 v Western Bulldogs
Gold Coast v Sydney +8
GWS +8 v Adelaide
Hawthorn +1 v Essendon
St Kilda +7 v Fremantle

6.5/9 from round 1.

Huh?

Your algorithm picked Port Adelaide to beat Brisbane comfortably in round 1 using the following form line since round 16 2022:

Port Adelaide: 3 wins (GWS, Ess, Adel) and 5 losses (Freo, Melb, Collingwood, Geelong, Richmond)

Brisbane: 7 wins (Doggies, GWS, GC, Carlton, St. Kilda, Richmond, Melb) and 4 losses (Ess, Richmond, Melb, Geelong)

 

[Fadge]

Hang on....is this the return of Roby and his magic beans?!?

 

[LordLucifer]

***** nerd-burger alert *****

 

[Hobbes]

Huh?

Your algorithm picked Port Adelaide to beat Brisbane comfortably in round 1 using the following form line since round 16 2022:

Port Adelaide: 3 wins (GWS, Ess, Adel) and 5 losses (Freo, Melb, Collingwood, Geelong, Richmond)

Brisbane: 7 wins (Doggies, GWS, GC, Carlton, St. Kilda, Richmond, Melb) and 4 losses (Ess, Richmond, Melb, Geelong)

Yep. It didn't like that Brisbane's most recent results included two hammerings (R23, Prelim), and most of the wins were small wins against lowly-rated teams. In comparison, Port finished 2023 with two thumping wins, and it didn't at all mind the small losses to Melbourne, Collingwood and Geelong.

Win-loss doesn't come into it much. And Port's home advantage is assessed at two goals, so an away loss smaller than that will be scored as a win. Also, the 10% deterioration of old ratings means that more recent results count more.

 

[Hobbes]

Retrospective tips - Round 2.

Ratings

Geelong 30.05
Sydney 19.74
Melbourne 17.62
Port Adelaide 14.64
Collingwood 14.27
Richmond 11.3
Fremantle 4.27
Carlton 3.51
Essendon 1.18
St Kilda -0.62
Brisbane -2.33
Adelaide -3.58
Western Bulldogs -3.76
West Coast -7.29
Gold Coast -7.54
GWS -7.64
Hawthorn -10.86
North Melbourne -13.94

Carlton v Geelong +27
Brisbane v Melbourne +14
Collingwood +12 v Port Adelaide
Adelaide v Richmond +3
Western Bulldogs v St Kilda +3
Fremantle +30 v North Melbourne
Sydney +37 v Hawthorn
Essendon +15 v Gold Coast
West Coast +12 v GWS

6/9, running total 12.5/18

 

[Fadge]

Interesting to see Collingwood's rating improved significantly between rounds 1 and 2, enough to see them favoured to beat Port Adelaide.

I smell something fishy...

 

[Hobbes]

Retrospective tips, round 3.

Ratings

Geelong 27.27
Sydney 21.23
Collingwood 20.32
Melbourne 15.19
Richmond 15.13
Port Adelaide 8.95
Carlton 6.88
St Kilda 6.12
Essendon 2.28
Brisbane 1.12
Fremantle -0.08
Adelaide -5.91
GWS -8.11
West Coast -8.11
Gold Coast -9.03
Western Bulldogs -9.42
North Melbourne -11.26
Hawthorn -14.85

Western Bulldogs v Brisbane +5
Collingwood +5 v Richmond
Hawthorn v North Melbourne +3
GWS v Carlton +9
St Kilda +4 v Essendon
Port Adelaide +15 v Adelaide
Gold Coast v Geelong +30
Melbourne v Sydney +1 (actually, the difference is 0.04)
Fremantle +8 v West Coast

4/9. Running total 16.5/27

 

[Hobbes]

Interesting to see Collingwood's rating improved significantly between rounds 1 and 2, enough to see them favoured to beat Port Adelaide.

I smell something fishy...

Collingwood beat the massively-rated Geelong in round 1. Their rating was always going up a lot.

In any case, with 12 points of home ground advantage coming into play, they were going to be tipped against Port regardless.

 

[Hobbes]

Retrospective tips, round 4.

Melbourne 20.63
Geelong 20.34
Collingwood 19.45
Sydney 16.07
Richmond 14.87
St Kilda 9.12
Carlton 6.92
Fremantle 4.4
Port Adelaide 2.6
Essendon 0.08
Adelaide -1.79
Brisbane -1.84
Gold Coast -3.83
GWS -7.1
Western Bulldogs -7.18
West Coast -11.15
Hawthorn -13.43
North Melbourne -14.69

Brisbane v Collingwood +15
North Melbourne v Carlton +22
Adelaide +6 v Fremantle
Richmond +22 v Western Bulldogs
St Kilda +19 v Gold Coast
Sydney +31 v Port Adelaide
Essendon +13 v GWS
West Coast v Melbourne +20
Geelong +34 v Hawthorn

6/9, running total 22.5/36