Total least squares

In applied statistics, total least squares is a type of errors-in-variables regression, a least squares data modeling technique in which observational errors on both dependent and independent variables are taken into account. It is a generalization of Deming regression and also of orthogonal regression, and can be applied to both linear and non-linear models.

The total least squares approximation of the data is generically equivalent to the best, in the Frobenius norm, low-rank approximation of the data matrix.

Linear model

Background

In the least squares method of data modeling, the objective function, ''S'',
:$S=\mathbf,$
is minimized, where ''r'' is the vector of residuals and ''W'' is a weighting matrix. In linear least squares the model contains equations which are linear in the parameters appearing in the parameter vector $\boldsymbol\beta$, so the residuals are given by
:$\mathbf.$
There are ''m'' observations in y and ''n'' parameters in β with ''m''>''n''. X is a ''m''×''n'' matrix whose elements are either constants or functions of the independent variables, x. The weight matrix W is, ideally, the inverse of the variance-covariance matrix $\mathbf M_y$ of the observations y. The independent variables are assumed to be error-free. The parameter estimates are found by setting the gradient equations to zero, which results in the normal equations
An alternative form is $\mathbf$, where $\boldsymbol\Delta \boldsymbol\beta$ is the parameter shift from some starting estimate of $\boldsymbol\beta$ and $\boldsymbol\Delta \mathbf y$ is the difference between y and the value calculated using the starting value of $\boldsymbol\beta$
:$\mathbf.$

Allowing observation errors in all variables

Now, suppose that both x and y are observed subject to error, with variance-covariance matrices $\mathbf M_x$ and $\mathbf M_y$ respectively. In this case the objective function can be written as
:$S=\mathbf,$
where $\mathbf r_x$ and $\mathbf r_y$ are the residuals in x and y respectively. Clearly these residuals cannot be independent of each other, but they must be constrained by some kind of relationship. Writing the model function as $\mathbf$, the constraints are expressed by ''m'' condition equations.

:$\mathbf.$
Thus, the problem is to minimize the objective function subject to the ''m'' constraints. It is solved by the use of Lagrange multipliers. After some algebraic manipulations, the result is obtained.

:$\mathbf,$

or alternatively $\mathbf,$
where M is the variance-covariance matrix relative to both independent and dependent variables.
:$\mathbf.$

Example

When the data errors are uncorrelated, all matrices M and W are diagonal. Then, take the example of straight line fitting.
:$f(x_i,\beta)=\alpha + \beta x_i$
in this case
:$M_=\sigma^2_+\beta^2 \sigma^2_$
showing how the variance at the ''i''th point is determined by the variances of both independent and dependent variables and by the model being used to fit the data. The expression may be generalized by noting that the parameter $\beta$ is the slope of the line.
:$M_=\sigma^2_+\left(\frac\right)^2_i \sigma^2_$

An expression of this type is used in fitting pH titration data where a small error on ''x'' translates to a large error on y when the slope is large.

Algebraic point of view

As was shown in 1980 by Golub and Van Loan, the TLS problem does not have a solution in general. The following considers the simple case where a unique solution exists without making any particular assumptions.

The computation of the TLS using singular value decomposition (SVD) is described in standard texts. We can solve the equation
:$XB \approx Y$
for ''B'' where ''X'' is ''m''-by-''n'' and ''Y'' is ''m''-by-''k''. The notation ''XB'' ≈ ''Y'' is used here to reflect the notation used in the earlier part of the article. In the computational literature the problem has been more commonly presented as ''AX'' ≈ ''B'', i.e. with the letter ''X'' used for the ''n''-by-''k'' matrix of unknown regression coefficients.

That is, we seek to find ''B'' that minimizes error matrices ''E'' and ''F'' for ''X'' and ''Y'' respectively. That is,
:\mathrm_ \, \; F\, _F, \qquad (X+E) B = Y+F
where \; F/math> is the augmented matrix with ''E'' and ''F'' side by side and $\, \cdot\, _F$ is the Frobenius norm, the square root of the sum of the squares of all entries in a matrix and so equivalently the square root of the sum of squares of the lengths of the rows or columns of the matrix.

This can be rewritten as
: X+E) \; (Y+F)\begin B\\ -I_k\end = 0.
where $I_k$ is the $k\times k$ identity matrix.
The goal is then to find \; F/math> that reduces the rank of \; Y/math> by ''k''. Define Sigma * to be the singular value decomposition of the augmented matrix \; Y/math>.
: \; Y= _X\; U_Y\begin\Sigma_X &0 \\ 0 & \Sigma_Y\end\beginV_ & V_ \\ V_ & V_\end^* = _X\; U_Y\begin\Sigma_X &0 \\ 0 & \Sigma_Y\end \begin V_^* & V_^* \\ V_^* & V_^*\end
where ''V'' is partitioned into blocks corresponding to the shape of ''X'' and ''Y''.

Using the Eckart–Young theorem, the approximation minimising the norm of the error is such that matrices $U$ and $V$ are unchanged, while the smallest $k$ singular values are replaced with zeroes. That is, we want
: X+E)\; (Y+F)= _X\; U_Y\begin\Sigma_X &0 \\ 0 & 0_\end\beginV_ & V_ \\ V_ & V_\end^*
so by linearity,
: \; F= - _X\; U_Y\begin0_ &0 \\ 0 & \Sigma_Y\end\beginV_ & V_ \\ V_ & V_\end^*.
We can then remove blocks from the ''U'' and Σ matrices, simplifying to
: \; F= -U_Y\Sigma_Y \beginV_\\V_\end^*= - \; Y\beginV_\\V_\end\beginV_\\ V_\end^*.
This provides ''E'' and ''F'' so that
: X+E) \; (Y+F)\beginV_\\ V_\end = 0.
Now if $V_$ is nonsingular, which is not always the case (note that the behavior of TLS when $V_$ is singular is not well understood yet), we can then right multiply both sides by $-V_^$ to bring the bottom block of the right matrix to the negative identity, giving
: X+E) \; (Y+F)\begin -V_ V_^ \\ -V_ V_^\end = X+E) \; (Y+F)\begin B\\ -I_k\end = 0 ,
and so
:$B=-V_ V_^.$

A naive GNU Octave implementation of this is:

function B = tls(X, Y)

n = size(X); % n is the width of X (X is m by n)
Z = Y % Z is X augmented with Y.
S V= svd(Z, 0); % find the SVD of Z.
VXY = V(1:n, 1+n:end); % Take the block of V consisting of the first n rows and the n+1 to last column
VYY = V(1+n:end, 1+n:end); % Take the bottom-right block of V.
B = -VXY / VYY;

end

The way described above of solving the problem, which requires that the matrix $V_$ is nonsingular, can be slightly extended by the so-called ''classical TLS algorithm''.

Computation

The standard implementation of classical TLS algorithm is available throug
Netlib
see also. All modern implementations based, for example, on solving a sequence of ordinary least squares problems, approximate the matrix $B$ (denoted $X$ in the literature), as introduced by Van Huffel and Vandewalle. It is worth noting, that this $B$ is, however, ''not the TLS solution'' in many cases.

Non-linear model

For non-linear systems

HOME


Content is Copyleft
Website design, code, and AI is Copyrighted (c) 2014-2017 by Stephen Payne


Consider donating to Wikimedia



As an Amazon Associate I earn from qualifying purchases