Certified Legendary Thread Race for the flag, in squiggly lines

[HoneyBadger35]

Oh my god, will everyone whose posts includes the letters V, F, and L in sequence, unless you are talking about historical squiggles, shut the **** up about it already!?!?

"I'm a good...work...guy?"
"You're fired."
"But I didn't say EEEEEEEEEEeeeeeeeeeeeeeeeeee."

 

[MWPP]

The fairest thing to do is to not have a premiership table.

And to keep this talk out of the Squiggle thread.

I was replying to another poster
Like me , you've helped keep this talk in the thread

 

[MWPP]

Anyway - back to the squiggle
I can't believe we're now higher than our 2006 premiership!

 

[The Speaker]

I can't believe we're now higher than our 2006 premiership!

i thought the club had cleaned itself up

 

[Smeg Head]

Anyway - back to the squiggle
I can't believe we're now higher than our 2006 premiership!

Oh s**t, look out here comes Walls.

 

[Glenno23]

Come here for squiggle. Leaving disappointed

 

[Final Siren]

Round 21, 2015​

Animated!

A top week of football, with upsets, a draw, thumpings, fear and consequences! Now we have two standout teams as well as a big gap between good teams and bad teams.

A super week for the Eagles, who flew into prime premiership territory. From here you can actually imagine them overtaking the Hawks. It needs a few more things to go their way... like they did on the weekend, with the Hawk juggernaut coming to a juddering halt against Port Adelaide.

The Power have never fallen that low this season, which is easy to overlook since high expectations make their season so disappointing. It's kind of the opposite of the St Kilda effect, where the Saints get praised for their improvement: It's there, but it's not enormous. The Power are still pretty good and the Saints are still pretty bad.

Also a good week for the Tigers, who added some much-needed attacking movement by dropping 23 goals on the Magpies. And Sydney moved a lot on the back of a shellacking of the Giants.

There are many candidates for worst week, but I think it has to go to the Bulldogs, who a week ago looked a sneaky chance to finish 4th and get a dream run with finals against Fremantle and Sydney, who the squiggle rates as far weaker than the Hawks and Eagles. It would have been a fairytale outcome for the Dogs, but after being thrashed by West Coast, they now face a likely elimination final against some in-form opposition (whoever that is).

Geelong had a shocker, too, as did Hawthorn, and Fremantle performed to expectation, i.e. badly. The squiggle is still really down on the Dockers. They're rated a little better than they were a few weeks ago, but in the meantime all their opposition have gotten better. I can't remember seeing a bigger gap between squiggle rating and ladder position.

Nine teams can't make the finals, but they can still finish the season strongly! In 2014, West Coast and Adelaide missed the finals but handed out some thumpings in the final rounds that caused them to start the 2015 squiggle in what appeared to be bizarrely high positions, and look how that turned out. Dead rubbers can be indicative of future performance.

On the ladder predictor - which really is no better than just tipping a few key games yourself, now we only have two rounds to go - we have Fremantle losing to Port while West Coast bag two more wins to steal the minor premiership. And Adelaide is favoured to make finals over the Cats on the balance of probabilities, even though this requires an upset somewhere, since the Crows are underdogs in their last two games while the Cats will start favourites.

Flagpole! Two teams and then a bunch of contenders.

I'll post more about flagpole below.

Live squiggle, which mostly worked well over the weekend with in-game updates!

 

[cats2rise]

I love the movement of Crows/Power/Roos/Dogs/Cats from Round 20 to Round 21. All are about to collide. COLLISION IS IMMINENT!

I assume this will be equivalent to the "Big Bang" in scientific terms. Please let it happen, Squiggle. Please.

 

[Final Siren]

Hi Final Siren,

Can you share the secret sauce behind the flagpole? There are a few things there that are a bit mysterious to me eg. teams which can't make the finals having a non-zero chance of winning the flag, and Adelaide, despite being rated 8th-best team by the squiggle and predicted to finish 8th are the 3rd favourite?

That would seem to suggest that recent performance are very heavily weighted, but (although I haven't been following the flagpole too closely), it doesn't seem to have the volatility that that would suggest.

I'd be very interested in how it's worked out.

Flagpole is purely a form rating and ignores where teams are on the ladder. It's like the regular squiggle in that all it knows is final scores of matches. (In fact, it doesn't even consider venue.) So what you're looking at is a rating from an algorithm that when fed all match results, has a good history of scoring the eventual premier highly.

What you suggest about weighting things so that Top 2 teams are higher and teams that can't make the finals are lower is sensible. The problem is I'm not sure I can do that better algorithmically than you can by thinking about it. So I don't do it at all.

It's very hard to isolate the benefit a team enjoys from finishing 2nd instead of 4th, or 4th instead of 5th. There's not much data and it's easy to think of plenty of reasons that any particular team under- or over-performed for any particular game.

Also, I think it's slightly interesting to see, for example, that Port Adelaide is a lot closer to a flag on exposed form than GWS, even though neither can make finals.

The maths work exactly like the regular squiggle ISTATE-91:12 algorithm, but it takes that crazy effect where a very strong defensive performance delivers big horizontal movement and adds it to the attacking plane as well, so that scoring way more than the squiggle expects will generate exponentially more movement.

It does weight recent results, but only because the ratings of good teams tend to inflate as the season progresses, so you wind up with high-rated teams where anyone who performs well against them also gets rated really highly. This means that flagpole positions don't usually change much for most of the season, but are increasingly volatile towards the back end.

That kind of thing is pretty cool, I think, because I didn't add that weighting factor; it just happens naturally when data runs through this simple algorithm.

That's basically my approach with this stuff: I let the computer brute-force all combinations of variables and algorithms against 20 years of match data, and then I have 250,000 algorithms that are hopeless tippers and one that's pretty good. So then I look at what that one is doing and try to figure out why it's good.

So far with flagpole I've deduced this kind of thing:

I'm still figuring out exactly why the flagpole algorithm seems to be good at what it does (picking premiers). It's only a little different to the regular squiggle but it plays out seasons quite differently.

The big difference I've noticed is it moves teams into better chart positions more easily but doesn't slide them back much. That is, it pays a lot of attention to great performances but almost ignores bad ones. There's very little penalty for putting a shocker vs a mediocre performance.

So the algorithm seems to be tracking a team's "upside," i.e. their potential best, rather than their average or typical performance like the normal squiggle does.

This effect also accelerates during the year, and by finals time some teams are on monstrous scores. Which means that performances against those teams then counts for even more!

In summary, from what I can tell so far, a good way to tip the premier is to:

• Rate unusually strong performances (including how many there are) but ignore bad games
• Rate strong performances against other strong teams
• Increasingly focus on late-season form
• Rate both strong attacking efforts and strong defensive efforts, even when they're not in tandem (e.g. a game where both teams score highly)

... most of which is probably obvious, but still.

Port Adelaide: They started very high and haven't been properly overtaken yet. Since this algorithm doesn't let teams slide much, it's basically not writing off Port's potential to deliver 2014-like performances yet. But they may well be left behind soon by teams that perform well against Hawthorn and West Coast.

Also the algorithm has liked Port's high scoring in their last two games (116 vs Adelaide, 129 vs Essendon).

But I'm still observing it. So far I think it's doing well. But if Fremantle win the flag this year, it will be a pretty savage blow to flagpole.

 

[Final Siren]

Final Siren - One thing I don't get.
Why does the forecast have North Melbourne winning 93-92 over the Bulldogs this week, but a 92-92 Elimination final?

I thought the algorithm just took home state into account. How does this get different results?

Aha, that's because I decided it was too stupid to tip a team by 0 points, so I tried to fix it, but did a half-arsed job.

The tipper works with decimals, so what it really does is decide North Melbourne will win by 17.152 points or whatever. Then it rounds that off, since otherwise it looks weird. Sometimes it thinks a team will win by less than half a point, and rounds off to zero.

So I changed that to bump it up to one point. But I forgot to make sure it happens in the finals tipper as well. I'll fix that.

 

[lusheslewis]

The inflation is an interesting effect. I suppose it could be "fixed" by starting teams on 0 attack/0 defense and adjusting the formula so that teams can get negative ratings. Have you tried this? (I get the feeling you probably have)

Maybe the inflation makes sense and is important to the system. You do expect teams to improve as the year goes on but ultimately it's just a tipping algorithm so only the relative ratings matter. Although I guess the flagpole is different in that respect (it's just trying to predict one outcome, not the winner of games each week).

Also the Eagle is in "No-Man's Land".

 

[Oneiros]

i though the club had cleaned itself up

Cracking shot.

 

[The Speaker]

But I'm still observing it. So far I think it's doing well. But if Fremantle win the flag this year, it will be a pretty savage blow to flagpole.

It would be, but right now Fremantle aren't demonstrating any of the qualities that you mention.

They haven't had unusually strong performances, all their losses have been against top 8 teams, their late season form is poor and their attack and defence isn't especially strong right now.

 

[Final Siren]

The inflation is an interesting effect. I suppose it could be "fixed" by starting teams on 0 attack/0 defense and adjusting the formula so that teams can get negative ratings. Have you tried this? (I get the feeling you probably have)

Maybe the inflation makes sense and is important to the system. You do expect teams to improve as the year goes on but ultimately it's just a tipping algorithm so only the relative ratings matter. Although I guess the flagpole is different in that respect (it's just trying to predict one outcome, not the winner of games each week).

Also the Eagle is in "No-Man's Land".

Right, until a few months ago I assumed that inflation was a flaw. The regular squiggle has a bit of it, too, which is why most teams finish the year higher than they started.

However, since it's clearly a major part of what makes flagpole work, I don't think it's a flaw any more. So I'm glad I didn't try to outsmart it.

 

[Chrizzt]

The Power have never fallen that low this season...

I get what you mean in terms of overall performance across the season, but given we lost to both Brisbane and Carlton I choose to snort in derision at the above statement.

 

[Glenno23]

WC have done an amazing job this year (and that pains me to say, living in WA as an SA expat is tough work! especially 05 and 06 )

They must be a good shot to take it all out, or will inexperience hurt them?

 

[djboonie]

WCE no threat as I reckon Hodgey should be able to bump WCE into that flagpole

 

[Gizza]

Could the Hawk line be in danger of finally being intersected?

 

[utility]

Could Freo be the lowest ever rated minor premier (going by the squiggle)?

Top of the ladder but look like they're fighting to make the finals.

 

[Roobs321]

I think the flagpole finally makes some sense this week from a visual standpoint, when you break the horizontal lines down into tiers; the hapless, the competitive, the capable, the contenders and the terrifying.

 

[treefingers]

Got a squiggle for commodities yet Final Siren ?

 

[swansfan51]

Swans squiggle resembles our gameplan

Stupid call. Our squiggle changes

 

[The King!]

Stupid call. Our squiggle changes

Visually though our ball movement is quite squiggly

 

[Rob]

We could really * the squiggle up this year.

 

[lusheslewis]

We could really **** the squiggle up this year.

If you win the flag from here you will probably move up to Geelong 2009/Sydney 2005 at least.