Faculty of Computers and Information
Faculty of Computers and Information
(Computer Exercise)
Department of Computer Science
Pattern Recognition
A First Simulation Example on Designing and Assessing a Regression Func
tion
•
Generate one small (10 observations) training dataset from binormal distribution (representing a response and a pre
dictor) with mean vector (0
,
0)
′
, unit variance, and
ρ
= 0
.
8
.
•
Fit the data to a linear model, calculate the apparent MSE given by
1
N
RSS
.
•
Plot the linear model, the data, and the best regression function (which is the conditional expectation of a bivariate
normal) on the same graph.
•
Generate a large data set (1000 observations to represent the population) from the same distribution and calculate the
true error rate (the MSE or the risk) of your model on this data set; we denote this by
err
tr
. This is the performance
conditional on the training set above. Obtain the performance of best regression function as well; denote it by
err
∗
.
This is the end of the preview.
Sign up
to
access the rest of the document.
 Spring '10
 WaleedA.Yousef
 Computer Science, Statistics, Regression Analysis, Probability theory, best regression function

Click to edit the document details