Scribd Logo
Upload

Welcome to Scribd!

  • 151 views

    Understanding Polyfitn

    Uploaded by

    Terry Ligard
    This document discusses and summarizes the polyfitn function, which performs polynomial regression. It solves for coefficients using linear least squares with a pivoted QR decomposition, which provides stable and numerically efficient estimates. Care is taken to scale variables and address ill-conditioning from nearly linearly dependent columns. The function takes independent and dependent variable values and a model specification as arguments.

    Copyright:

    © All Rights Reserved
    Download as RTF, PDF, TXT or read online from Scribd

    Understanding Polyfitn

    Uploaded by

    Terry Ligard
    151 views4 pages
    This document discusses and summarizes the polyfitn function, which performs polynomial regression. It solves for coefficients using linear least squares with a pivoted QR decomposition, which provides stable and numerically efficient estimates. Care is taken to scale variables and address ill-conditioning from nearly linearly dependent columns. The function takes independent and dependent variable values and a model specification as arguments.

    Original Description:

    This document helps you understand polyfitn

    Copyright

    © © All Rights Reserved

    Available Formats

    RTF, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses and summarizes the polyfitn function, which performs polynomial regression. It solves for coefficients using linear least squares with a pivoted QR decomposition, which provides stable and numerically efficient estimates. Care is taken to scale variables and address ill-conditioning from nearly linearly dependent columns. The function takes independent and dependent variable values and a model specification as arguments.

    Copyright:

    © All Rights Reserved
    Download as RTF, PDF, TXT or read online from Scribd
    Download as rtf, pdf, or txt
    151 views4 pages

    Understanding Polyfitn

    Uploaded by

    Terry Ligard
    This document discusses and summarizes the polyfitn function, which performs polynomial regression. It solves for coefficients using linear least squares with a pivoted QR decomposition, which provides stable and numerically efficient estimates. Care is taken to scale variables and address ill-conditioning from nearly linearly dependent columns. The function takes independent and dependent variable values and a model specification as arguments.

    Copyright:

    © All Rights Reserved
    Download as RTF, PDF, TXT or read online from Scribd
    Download as rtf, pdf, or txt
    of 4

    Understanding Polyfitn

    John R. D'Errico
    woodchips@rochester.rr.com

    8/8/06

    Regression modeling

    Polyfitn is really rather simple. It solves for the coefficients of a


    polynomial regression model using traditional linear least squares
    techniques, with some care applied in the solution. The method chosen
    for its solution is one that also yields standard errors for the
    coefficients.

    A good reference for regression analysis is the one I learned from:

    Draper, N.R., Smith, H.; "Applied Regression Analysis"; John Wiley &
    Sons

    Care has been taken in the numerical methods of polyfitn, first by use
    of a QR factorization with pivoting to solve the system. This is more
    stable than the simple, unpivoted QR. I've also used automatic variable
    scaling to deal with a simple cause of ill-conditioning.

    Numerical details of polyfitn

    Once the polynomial (or multinomial in n-dimensions) model has been


    specified, the estimation procedure is itself rather simple. The problem
    becomes that of estimation of the vector x, given the linear system of
    equations

    A*x = y

    Of course, for this estimation to have a unique solution, the matrix A


    should be both nonsingular and have more rows than columns.
    Problems with fewer rows than columns are called underdetermined.
    I'd strongly recommend that you get more data if there are fewer data
    points than parameters to estimate.
    So, assuming that A is an nxp matrix, with n>p, we can solve this
    system via many different approaches in Matlab. I'll outline the method
    I chose, as well as what I've done to assure the the best parameter
    estimates possible.

    The traditional methods in Matlab that one might use to estimate a


    linear regression model are

    - Backslash (\)
    - PINV
    - LSQR
    - The normal equations, i.e., x = (A' * A)\(A' * y)
    - QR decomposition
    - Pivoted QR decomposition

    Of these methods, only those based on the QR factorization will also


    directly yield estimated variances for the parameters. A pivoted QR is
    also reasonably efficient, as well as numerically stable, so I chose this
    method.

    Given the design matrix A in the above expression, the first task is to
    scale it so that all columns have unit mean. While I'd like to introduce
    a shift also for the best possible estimates, this is not possible for a
    general multinomial model, as it may introduce additional terms into
    the model. For example, suppose the polynomial model we must
    estimate is

    y = a1*x + a2*x^2

    If we translate the columns by automatically subtracting the mean,


    then we have implicitly inserted a constant term into the model which
    was not requested. So all that is done is a scaling of the matrix A.
    Essentially, we solve the problem

    A*D*inv(D)*x = y

    Here D is the diagonal matrix

    D = diag(1./mean(A,1));
    If we write

    z = inv(D)*x

    then the scaling results in the problem

    (A*D)*z = y

    Solve for z, then scale back to recover x.

    The next problem to worry about is the existence of nearly linearly


    dependent columns in A. I'll use an extreme example to illustrate what
    might happen otherwise. Consider an under-determined linear
    regression problem:

    n = 50;
    X = randn(n,3);
    X = X(:,[1 2 2 3]);

    Note that the third column is actually just a copy of the second
    column. Now, generate Y as the simple sum of columns [1 2 4] of X.

    Y = X*[1;1;0;1] + randn(n,1)/100000;

    The coefficients of our true model are just

    Y = 1*x1 + 1*x2 + 0*x3 + 1*x4

    Can we estimate

    Details of the arguments for polyfitn

    There are only three arguments to polyfitn.

    - An array (or vector) of independent variable values. Each column of


    this array corresponds to an independent variable. If a vector,
    - A vector of dependent variable values.

    - A model specification

    You might also like

    pFad - Phonifier reborn

    Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

    Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


    Alternative Proxies:

    Alternative Proxy

    pFad Proxy

    pFad v3 Proxy

    pFad v4 Proxy