Re: 2018 AFL Ladder predictions
Posted: Thu Feb 22, 2018 10:11 am
Lightning McQueen wrote::(
I'll leave the room then, sorry.
There's more people who think as you do than how I do mate.
Lightning McQueen wrote::(
I'll leave the room then, sorry.
Booney wrote:Whilst I haven't put it in my ladder I've got a funny gut feel about the Bulldogs and Richmond and I've got nothing to back it up other than the feeling in my waters.
Dogs to rise up, Richmond to fall over....I don't think it will happen but it's a thought I keep having when looking at the year ahead.
MW wrote:Melbourne are shit.
Booney wrote:Lightning McQueen wrote::(
I'll leave the room then, sorry.
There's more people who think as you do than how I do mate.
Lightning McQueen wrote:Booney wrote:Lightning McQueen wrote::(
I'll leave the room then, sorry.
There's more people who think as you do than how I do mate.
That scares me, I know what goes on in my head and some of it just aint right.
Booney wrote:
I meant about Melbourne mate, not riding a ladies bike with a carton in the basket.
Nice strawman, Benny.bennymacca wrote:morell wrote:its not measuring finals wins, that's fine, it's that 2014 is a completely arbitrary line in the sand. Why not pluck how many finals we have won since 2012 instead? If you're going to use "finals won since 2014" as a measurement, then you need to do that to all clubs and benchmark it.
How many finals did Richmond win in the three years before last year?
See? It's meaningless.
You blokes need to take a stats course.
so you think that a team's strength last year is in no way correlated to the strength of the team this year? (as well as many other factors)
never heard of bayes theorem?
you may need to take a stats course to learn about it
morell wrote:Nice strawman, Benny.bennymacca wrote:morell wrote:its not measuring finals wins, that's fine, it's that 2014 is a completely arbitrary line in the sand. Why not pluck how many finals we have won since 2012 instead? If you're going to use "finals won since 2014" as a measurement, then you need to do that to all clubs and benchmark it.
How many finals did Richmond win in the three years before last year?
See? It's meaningless.
You blokes need to take a stats course.
so you think that a team's strength last year is in no way correlated to the strength of the team this year? (as well as many other factors)
never heard of bayes theorem?
you may need to take a stats course to learn about it
Bayesian networks are handy for predicting probability when the conditions of the data are small. Soccer, is a good one, because you have limited scoring. Baseball, too. You can see upwards of a 70% accuracy for those sports. AFL? Where there are 18 players on the field, a oval ball, high scoring variables and shall we say... a unique interpretive set of rules means that method simply won't work too well for our sport.
But! That's not my point. Say you do want to use a Bayesian model to predict the probable strength of a team next year, you would have to select a statistically significant time frame to measure for and not just cherry pick, like old mate did, an arbitrary point in time, and then not benchmark that score against other samples.
Otherwise, you can select and manipulate the data to say almost anything you like.
The Bedge wrote:Why does everything need to transcend into a scientific/mathematical argument?
Why can't we just say shit coz that's what we think
Bombers4EVA wrote:morell wrote:Nice strawman, Benny.bennymacca wrote:morell wrote:its not measuring finals wins, that's fine, it's that 2014 is a completely arbitrary line in the sand. Why not pluck how many finals we have won since 2012 instead? If you're going to use "finals won since 2014" as a measurement, then you need to do that to all clubs and benchmark it.
How many finals did Richmond win in the three years before last year?
See? It's meaningless.
You blokes need to take a stats course.
so you think that a team's strength last year is in no way correlated to the strength of the team this year? (as well as many other factors)
never heard of bayes theorem?
you may need to take a stats course to learn about it
Bayesian networks are handy for predicting probability when the conditions of the data are small. Soccer, is a good one, because you have limited scoring. Baseball, too. You can see upwards of a 70% accuracy for those sports. AFL? Where there are 18 players on the field, a oval ball, high scoring variables and shall we say... a unique interpretive set of rules means that method simply won't work too well for our sport.
But! That's not my point. Say you do want to use a Bayesian model to predict the probable strength of a team next year, you would have to select a statistically significant time frame to measure for and not just cherry pick, like old mate did, an arbitrary point in time, and then not benchmark that score against other samples.
Otherwise, you can select and manipulate the data to say almost anything you like.
****.... What a brain fart...Lost interest after reading Bayesian.
morell wrote:Nice strawman, Benny.bennymacca wrote:morell wrote:its not measuring finals wins, that's fine, it's that 2014 is a completely arbitrary line in the sand. Why not pluck how many finals we have won since 2012 instead? If you're going to use "finals won since 2014" as a measurement, then you need to do that to all clubs and benchmark it.
How many finals did Richmond win in the three years before last year?
See? It's meaningless.
You blokes need to take a stats course.
so you think that a team's strength last year is in no way correlated to the strength of the team this year? (as well as many other factors)
never heard of bayes theorem?
you may need to take a stats course to learn about it
Bayesian networks are handy for predicting probability when the conditions of the data are small. Soccer, is a good one, because you have limited scoring. Baseball, too. You can see upwards of a 70% accuracy for those sports. AFL? Where there are 18 players on the field, a oval ball, high scoring variables and shall we say... a unique interpretive set of rules means that method simply won't work too well for our sport.
But! That's not my point. Say you do want to use a Bayesian model to predict the probable strength of a team next year, you would have to select a statistically significant time frame to measure for and not just cherry pick, like old mate did, an arbitrary point in time, and then not benchmark that score against other samples.
Otherwise, you can select and manipulate the data to say almost anything you like.
Booney wrote:The Bedge wrote:Why does everything need to transcend into a scientific/mathematical argument?
Why can't we just say shit coz that's what we think
I would have said the last couple of years has little to do with this year because we've turned over 11 players in our squad, but that would be too simplistic. I think.
bennymacca wrote:morell wrote:Nice strawman, Benny.bennymacca wrote:morell wrote:its not measuring finals wins, that's fine, it's that 2014 is a completely arbitrary line in the sand. Why not pluck how many finals we have won since 2012 instead? If you're going to use "finals won since 2014" as a measurement, then you need to do that to all clubs and benchmark it.
How many finals did Richmond win in the three years before last year?
See? It's meaningless.
You blokes need to take a stats course.
so you think that a team's strength last year is in no way correlated to the strength of the team this year? (as well as many other factors)
never heard of bayes theorem?
you may need to take a stats course to learn about it
Bayesian networks are handy for predicting probability when the conditions of the data are small. Soccer, is a good one, because you have limited scoring. Baseball, too. You can see upwards of a 70% accuracy for those sports. AFL? Where there are 18 players on the field, a oval ball, high scoring variables and shall we say... a unique interpretive set of rules means that method simply won't work too well for our sport.
But! That's not my point. Say you do want to use a Bayesian model to predict the probable strength of a team next year, you would have to select a statistically significant time frame to measure for and not just cherry pick, like old mate did, an arbitrary point in time, and then not benchmark that score against other samples.
Otherwise, you can select and manipulate the data to say almost anything you like.
Not sure you understand what bayes theorem is. All bayes theorem does is allow you to modify the probabilities based on a priori knowledge.
Or in simple terms we can update our base assumptions by adding in things we know about the system we are modelling. You don’t have to know anything about the scoring, it’s irrelevant in this case, we are modelling team strength.
A base assumption might be that every team has an equal chance of winning the flag. But due to information we can gather - like recent results - we can update our estimates.
So in that case the fact that port hasn’t won any recent games against good opposition is absolutely relevant. With more recent results weighting more heavily than results further into the past.
Exactly how much we weight this information isn’t clear. But it certainly isn’t irrelevant
Booney wrote:The Bedge wrote:Why does everything need to transcend into a scientific/mathematical argument?
Why can't we just say shit coz that's what we think
I would have said the last couple of years has little to do with this year because we've turned over 11 players in our squad, but that would be too simplistic. I think.
MW wrote:By this theory are you going to stop mentioning how long it's been since Adelaide won a GF since it's irrelevant to this years competition? I didn't think so