Certified Legendary Thread Race for the flag, in squiggly lines

[Epilegs]

Dammit, Hawthorn! Come back!

 

[The Speaker]

Max, any chance we could get some WAFL/SANFL squiggles one day?

 

[barmy44]

Love this thread FS! Been great watching Port's slow trajectory upwards. Obviously still a long way to go to bridge the gap between the three best teams in the comp: Hawthorn, Freo and Geelong are all out by quite a bit compared to the rest.

i have a feeling port will be in the leading pack sooner rather than later...

i also have a feeling wingard will be the GAJ of his 'generation'.

 

[Leeda]

and the lord raises Geelong from the dead just for pernicious purposes... no I don't know what it means but I said it anyway.

 

[Air_Chester]

Here comes Essendon.

I just so badly want to make a premature ejaculation joke about the current Squiggle status. So badly. But instead I'll just do watch any bloke with that issue probably does: try to enjoy it while it lasts.

Oh wait.

 

[Billy ray]

I know these are just visualisations, but so far for the year, how have the tips gone?

 

[Final Siren]

Max, any chance we could get some WAFL/SANFL squiggles one day?

If I have the data, I can generate a squiggle for it no problems. But it would probably take a lot of effort to get good WAFL/SANFL data, and parse it into the right form. For AFL, a little bot runs out on the internet and gathers it for me.

 

[Leather Poisoning]

Was really concerned about the run of the Power but apparently we'll win 96 - 83.

Will sleep easier now!!!! We win Sunday and this is thread of the year!!

97-90 good enough for ya?

 

[Final Siren]

I now these are just visualisations, but so far for the year, how have the tips gone?

17/27 for the squiggle's algorithm of choice, which goes by the pet name of ISTATE-91:12. It's currently ranked #31 out of my 92 algorithms, behind HOMULT-80:16 on 20/27. That's a similar algorithm that uses a multiplier for home ground advantage instead of a constant... a 16% bonus/penalty to scores. It also awards home ground advantage for every game, not just ones involving a team travelling interstate. HOMULT-80:16 also adapts more quickly to new data (weighting previous results at 80% rather than 91%), which probably helps early in the year. I'm sure it's mostly just lucky so far, though, since it's a middling performer over the long term.

The ISTATE-based algorithms are having a bit of a rough patch; the results haven't been great for interstate home advantage.

 

[philohk]

17/27 for the squiggle's algorithm of choice, which goes by the pet name of ISTATE-91:12. It's currently ranked #31 out of my 92 algorithms, behind HOMULT-80:16 on 20/27. That's a similar algorithm that uses a multiplier for home ground advantage instead of a constant... a 16% bonus/penalty to scores. It also awards home ground advantage for every game, not just ones involving a team travelling interstate. HOMULT-80:16 also adapts more quickly to new data (weighting previous results at 80% rather than 91%), which probably helps early in the year. I'm sure it's mostly just lucky so far, though, since it's a middling performer over the long term.

The ISTATE-based algorithms are having a bit of a rough patch; the results haven't been great for interstate home advantage.

areyouawizard.jpg

Seriously, excellent work and genuinely fascinating, FS - thanks for sharing.

 

[NarniMagic]

17/27 for the squiggle's algorithm of choice, which goes by the pet name of ISTATE-91:12. It's currently ranked #31 out of my 92 algorithms, behind HOMULT-80:16 on 20/27. That's a similar algorithm that uses a multiplier for home ground advantage instead of a constant... a 16% bonus/penalty to scores. It also awards home ground advantage for every game, not just ones involving a team travelling interstate. HOMULT-80:16 also adapts more quickly to new data (weighting previous results at 80% rather than 91%), which probably helps early in the year. I'm sure it's mostly just lucky so far, though, since it's a middling performer over the long term.

The ISTATE-based algorithms are having a bit of a rough patch; the results haven't been great for interstate home advantage.

So now I'm curious... What's your top ranked algorithm?

 

[The Speaker]

If I have the data, I can generate a squiggle for it no problems. But it would probably take a lot of effort to get good WAFL/SANFL data, and parse it into the right form. For AFL, a little bot runs out on the internet and gathers it for me.

I'm thinking more historically than currently - does wherever you got your VFL data from also have WAFL/SANFL data?

 

[philohk]

So now I'm curious... What's your top ranked algorithm?

I think he means it's HOMULT, with 20/27.

 

[Final Siren]

areyouawizard.jpg

Seriously, excellent work and genuinely fascinating, FS - thanks for sharing.

Ha, don't tempt me... ahhh too late.

Here are some basic algorithms and their performance over the last 20 years!

RANDOM, which works like you would expect, gets about 50% right. Or slightly less (stupid draws).

MOMENTUM, tipping whoever won last week and tie-breaking on biggest winning margin, is right 57.7% of the time.

HOMER, always tipping the home team, is 60.1%.

PERCY, tipping the team with the best percentage, is 63.2%.

MOREWINS, tipping the team with the most wins and tie-breaking on home team, is 65.5%.

Any better than that requires more complicated algorithms!

ISTATE-90:12, the one I use for squiggles, is 67.2% accurate over the last 20 years. It's been quite a lot stronger over the last ten years (69.2%), though, and particularly the last five years (72.9%), which is why I use it.

ISTATEVH-91:12 is slightly more accurate (67.9% over 20 years, which is the third best algorithm over that time period, and 73.3% over the last 5 years, which is the best), but more complicated, so I decided the extra complexity wasn't worth it for squiggles. It verifies home ground advantage a little more rigorously and awards a small home ground advantage to home teams in neutral venues.

The algorithm I find most interesting though is SHOTS, which makes up its own scoring system. The idea is that although kicking a behind rather than a goal is worth a lot less to your score (one sixth as much, in fact), it's not necessarily that much of a stronger indicator as to who's a better team. It might just be that your kicking was a little off that day. So SHOTS-5:2:ISTATEVH:87:11 counts goals as if they were worth 5 points and behinds 2 points and runs those numbers through the ISTATEVH algorithm. And it does a pretty good job! That one is running at 72.4% accuracy over the last five years (my 5th best algorithm for that time period), and 67.9% over the last 20 years (2nd best).

In fact, my best algorithm over the last 20 years is SHOTS-3:3:ISTATEVH:91:12, which awards, yes, 3 points for a goal and 3 points for a behind. I like it because it implies that generating a lot of shots on goal is a very good indicator of team strength, regardless of how many shots are goals.

Sadly, SHOTS-3:3 isn't as good over the last five years, only tipping at 70.0% (26th overall).

 

[cryptor]

Sadly, SHOTS-3:3 isn't as good over the last five years, only tipping at 70.0% (26th overall).

Do you have a theory why that is? Something to do with how the game is being played now?

 

[SunshineBoy]

Thread needs more graphs

 

[The 747]

areyouawizard.jpg

Seriously, excellent work and genuinely fascinating, FS - thanks for sharing.

x2. This is a really interesting thread, well done.

 

[tazaa]

Wiz.

 

[KillaKanga]

97-90 good enough for ya?

Bloody oath

Sent from my GT-I9305T using Tapatalk

 

[Simon_Nesbit]

....
In fact, my best algorithm over the last 20 years is SHOTS-3:3:ISTATEVH:91:12, which awards, yes, 3 points for a goal and 3 points for a behind. I like it because it implies that generating a lot of shots on goal is a very good indicator of team strength, regardless of how many shots are goals.

Sadly, SHOTS-3:3 isn't as good over the last five years, only tipping at 70.0% (26th overall).

Amazing.

I remember Clarkson commenting before the 07(?) season that the clubs analysis had worked out that scoring accuracy was one of the least influential performance markers -there was a small WTF? reaction, then the comment disappeared into the background.

I wonder what other statistical modelling the clubs use, and how much this influences their training and performances.

 

[Final Siren]

Round 4, 2014

Play with dynamic squiggles

 

[Heaps of fun]

North making a move.

 

[autocol]

Melbourne went upwards! On both measures! Hallelujah!

Also, Final Siren you are my favourite Bigfooty poster.

 

[Seppo]

Final Siren, have you run an algorithm for ave games experience? Wonder if that hit rate would be closer to the mark than Shots or Home ground.

 

[Dazb86]

Stuff the ladder. Dogs v Carlton are battling for squiggle supremacy this week.