COS 323, Fall 2010, Assignment 3
README.txt

Name:
NetID:


General Questions
-----------------

1. Find the best (least-squares) quadratic approximation
(that is, y = a + b*x + c*x^2) to y = e^x on the interval x = [-1, 1].
You may find the answer using any combination of analytic and
numerical methods you wish.


[your answer here]


2. The Hilbert matrix of order n is a square matrix in which
the ij-th entry has value 1/(i+j-1).  You can produce one
using Matlab's "hilb" function.  Use the svd command in Matlab
to find the condition number of the Hilbert matrix of size 4.
(The Hilbert matrices are known for, among other things, having
huge condition numbers.)


[your answer here]


3. Consider the overconstrained system of equations
  2x +  y = 2
   x + 2y = 1
  2x - 2y = -1

a) Find the total-least-squares solution.
b) Find the best solution(s) under the 1-norm (i.e., the solution
minimizing the sum of absolute values of distances to the lines).
c) Find the best solution under the infinity-norm (i.e., the
solution minimizing the *maximum* distance to any of the lines).

Hint: For parts (b) and (c), draw yourself a picture and think
geometrically - don't embark on IRLS or the like.


[your answer here]



Programming assignment: Logistic Regression
-------------------------------------------

List the optimal coefficients you found for gradschool.csv
and how stable they were for different subsets of data.


[your answer here]


Brief description of your code, termination criterion, and any notes:


[your answer here]