network profile
| bio | website | |
|---|---|---|
| location | ||
| age | ||
| visits | member for | 1 year |
| seen | May 6 '12 at 16:23 | |
| stats | profile views | 2 |
|
Apr 30 |
comment |
About Akaike's criterion and VC-dimension of linear regressors I again thank you, but here is a quote from "Multimodel Inference: Understanding AIC and BIC in Model Selection", by K.P.Burnham and D.R.Anderson,Sociological Methods Research 2004; 33; 261 "Akaike found that the maximized log-likelihood value was a biased estimate of EyEx[log(g(x| ˆ θ(y)))], but this bias was approximately equal to K, the number of estimable parameters in the approximating model, g (for details, see Burnham and Anderson 2002, chap. 7). This is an asymptotic result of fundamental importance." |
|
Apr 30 |
awarded | Student |
|
Apr 30 |
comment |
About Akaike's criterion and VC-dimension of linear regressors Thanks for trying, but your answer is of little help. The question I'm asking is: say we have a ML estimate of a linear regressor, and let R_emp be the residual of this estimate. Is such residual a biased or unbiased estimate of the generalization error of the ML regressor? VC-theory says it is asymptotically unbiased. If so, Akaike's criterion cannot be justified as it is usually done. Thanks again |
|
Apr 30 |
comment |
About Akaike's criterion and VC-dimension of linear regressors Thanks for trying, but your answer is of little help. The question I'm asking is: say we have a ML estimate of a linear regressor, |
|
Apr 29 |
asked | About Akaike's criterion and VC-dimension of linear regressors |
