Home
Articles
Photos
Interviews
Forums
New posts
Search forums
Georgia Tech Recruiting
Dashboard
What's new
New posts
New profile posts
Latest activity
Chat
Log in
Register
What's new
Search
Search
Search titles only
By:
New posts
Search forums
Menu
Log in
Register
Install the app
Install
Home
Forums
Georgia Tech Athletics
Georgia Tech Football
Stakes even higher in ACC-SEC rivalry
JavaScript is disabled. For a better experience, please enable JavaScript in your browser before proceeding.
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an
alternative browser
.
Reply to thread
Message
<blockquote data-quote="DTGT" data-source="post: 105438" data-attributes="member: 455"><p>Models that are line plus noise have consistently been inferior to the opening line and updated line. That model may correlate with the line and the line may correlate with the result, but that doesn't mean that the model will correlate to the result.</p><p></p><p>Example: Jar of beans.</p><p>1. Have a crowd guess the number of beans in a jar. The average of the guesses will correlate with the number of beans in the jar.</p><p>2. Develop a model that will estimate the average crowd guess based on jar and type of bean. Ignore the actual number of beans in the jar. Error is the difference between the estimate and the crowd guess.</p><p>3. Model is "improved" by reducing the error between the model's guess and the crowd's guess.</p><p>4. Error between model and crowd exists. There is now error from jar to crowd and crowd to model. Error from the model to the jar is necessarily worse than the error from the crowd to the jar (this is essentially by definition of how errors add and accumulate).</p><p></p><p>In our above example, any systemic error that is present in the crowd guess will be systemic in the model of the crowd guess. Per <a href="http://www.atomicfootball.com/" target="_blank">http://www.atomicfootball.com/</a> the mean square error of the line is about 235 (root mean square = 15.3) and the mean absolute error is about 12. That the root mean square error is 25% greater than the mean absolute error shows that the line has at least some bias in one direction. What direction? Indeterminate. If you could determine the bias, you could become very rich.</p><p></p><p>True Error = Absolute Error + Bias</p><p></p><p>Bias in the line will be transferred to the model. Additionally, modeling the line will add its own absolute error and its own bias error. <strong>If you make a model of a model of a cat, you don't have a cat; you have a worse model of a cat.</strong></p></blockquote><p></p>
[QUOTE="DTGT, post: 105438, member: 455"] Models that are line plus noise have consistently been inferior to the opening line and updated line. That model may correlate with the line and the line may correlate with the result, but that doesn't mean that the model will correlate to the result. Example: Jar of beans. 1. Have a crowd guess the number of beans in a jar. The average of the guesses will correlate with the number of beans in the jar. 2. Develop a model that will estimate the average crowd guess based on jar and type of bean. Ignore the actual number of beans in the jar. Error is the difference between the estimate and the crowd guess. 3. Model is "improved" by reducing the error between the model's guess and the crowd's guess. 4. Error between model and crowd exists. There is now error from jar to crowd and crowd to model. Error from the model to the jar is necessarily worse than the error from the crowd to the jar (this is essentially by definition of how errors add and accumulate). In our above example, any systemic error that is present in the crowd guess will be systemic in the model of the crowd guess. Per [url]http://www.atomicfootball.com/[/url] the mean square error of the line is about 235 (root mean square = 15.3) and the mean absolute error is about 12. That the root mean square error is 25% greater than the mean absolute error shows that the line has at least some bias in one direction. What direction? Indeterminate. If you could determine the bias, you could become very rich. True Error = Absolute Error + Bias Bias in the line will be transferred to the model. Additionally, modeling the line will add its own absolute error and its own bias error. [B]If you make a model of a model of a cat, you don't have a cat; you have a worse model of a cat.[/B] [/QUOTE]
Insert quotes…
Verification
What is the last name of the current Head Football Coach?
Post reply
Home
Forums
Georgia Tech Athletics
Georgia Tech Football
Stakes even higher in ACC-SEC rivalry
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
Accept
Learn more…
Top