restrSigmaGPCM

restrSigmaGPCM computes constrained covariance matrices for the 14 GPCM specifications

expand all in page

Syntax

  • Sigma=restrSigmaGPCM(SigmaB, niini, pa, nocheck)example
  • Sigma=restrSigmaGPCM(SigmaB, niini, pa, nocheck, lmd)example
  • Sigma=restrSigmaGPCM(SigmaB, niini, pa, nocheck, lmd, OMG)example
  • [Sigma, lmd]=restrSigmaGPCM(___)example
  • [Sigma, lmd, OMG]=restrSigmaGPCM(___)example
  • [Sigma, lmd, OMG, GAM]=restrSigmaGPCM(___)example

Description

This routine applies the constraints to the covariance matrices using the specifications contained in input structure pa.

example

Sigma =restrSigmaGPCM(SigmaB, niini, pa, nocheck) Use of restrSigmaGPCM with just one output argument.

example

Sigma =restrSigmaGPCM(SigmaB, niini, pa, nocheck, lmd) Use of restrSigmaGPCM with all default options.

example

Sigma =restrSigmaGPCM(SigmaB, niini, pa, nocheck, lmd, OMG) Use function genSigmaGPCM to generate the covariance matrices.

example

[Sigma, lmd] =restrSigmaGPCM(___) Generate ad hoc cov matrices.

example

[Sigma, lmd, OMG] =restrSigmaGPCM(___)

example

[Sigma, lmd, OMG, GAM] =restrSigmaGPCM(___)

Examples

expand all

  • Use of restrSigmaGPCM with just one output argument.

    • Generate 3 cov matrices of size 2-by-2 

    rng('default')
rng(1500)
Sig = [1 .5; .5 2];
df = 10;
k=2;
v=2;
Sigma=zeros(v,v,k);
for j=1:k
Sigma(:,:,j) = wishrnd(Sig,df)/df;
end
niini=2;
% Apply model EEE.
pa=struct;
pa.pars='EEE';
SigmaNEW  = restrSigmaGPCM(Sigma, niini, pa);

  • Use of restrSigmaGPCM with all default options.

    • Generate 3 cov matrices of size 2-by-2 

    rng('default')
rng(1500)
Sig = [1 .5; .5 2];
df = 10;
k=3;
v=2;
Sigma=zeros(v,v,k);
for j=1:k
Sigma(:,:,j) = wishrnd(Sig,df)/df;
end
niini=[20 150 200];
% Apply model VVE.
pa=struct;
pa.pars='VVE';
[SigmaNEW, lmd, OMG, GAM]  = restrSigmaGPCM(Sigma, niini, pa);

  • Use function genSigmaGPCM to generate the covariance matrices.

    • v=3;
k=5;
pa=struct;
pa.pars='EVI';
S=genSigmaGPCM(v, k, pa);
niini=100*ones(1,k);
pa.pars='VVE';
[SigmaNEW, lmd, OMG, GAM]  = restrSigmaGPCM(S, niini, pa);

  • Generate ad hoc cov matrices.

    • k=7; v=20; n=100;
rng('default')
seed=1141;
add=ones(v,v)+diag(1:v);
niini= round(100*mtR(k,0,seed));
Sigma=zeros(v,v,k);
for j=1:k
Sigma(:,:,j)=cov(reshape(mtR(n*v,1,-1),n,v)).*add;
end
sph=struct;
sph.pars='EVV';
niini=100*ones(k,1);
[SigmaNEW, lmd, OMG, GAM]  = restrSigmaGPCM(Sigma, niini', sph);

    Input Arguments

    expand all

    SigmaB — initial unconstrained covariance matrices. p-by-p-by-k array.

    p-by-p-by-k array containing the k covariance matrices for the k groups.

    Data Types: single|double

    niini — sizes of the groups. Vector.

    Vector of length k containing the size of the groups. ninini can be either a row or a columns vector.

    Data Types: single|double

    pa — Constraints to apply and model specification. Structure.

    Structure containing the following fields:

    Value Description
    pars

    type of Gaussian Parsimonious Clustering Model. Character.

    A 3 letter word in the set: 'VVE','EVE','VVV','EVV','VEE','EEE','VEV','EEV','VVI', 'EVI','VEI','EEI','VII','EII'.

    The field pa.pars is compulsory. All the other fields are non necessary. If they are not present they are set to their default values.
    cdet

    scalar in the interval [1 Inf) which specifies the the restriction which has to be applied to the determinants.

    If pa.cdet=1 all determinants are forced to be equal.

    See section More About for additional details.
    shw

    scalar in the interval [1 Inf) which specifies the the restriction which has to be applied to the elements of the shape matrices inside each group. If pa.shw=1 all diagonal elements of the shape matrix of cluster j (with j=1, ..., k) will be equal.
    shb

    scalar in the interval [1 Inf) which specifies the the restriction which has to be applied to the elements of the shape matrices across each group.
    maxiterS

    positive integer which specifies the maximum number of iterations to obtain the restricted shape matrix.

    This parameter is used by routine restrshapepars. The default value of pa.maxiterS is 5.
    maxiterR

    positive integer which specifies the maximum number of iterations to obtain the common rotation matrix in presence of varying shape.

    This parameter is used by routine cpcV. The default value of pa.maxiterR is 20.
    maxiterDSR

    positive integer which specifies the maximum number of iterations to obtain the requested restricted determinants, shape matrices and rotation. For all parametrizations pa.maxiterDSR is set to 1 apart from for the specifications 'VVE', 'EVE' and 'VEE'. The default value of pa.maxiterDSR is 20.
    tolS

    tolerance to use to exit the iterative procedure for estimating the shape. Scalar. The iterative procedures stops when the relative difference of a certain output matrix is smaller than itertol in two consecutive iterations. The default value of pa.tol is 1e-12.
    zerotol

    tolerance value to declare all input values equal to 0 in the eigenvalues restriction routine (file restreigen.m) or in the final reconstruction of covariance matrices.

    The default value of zerotol is 1e-10.
    msg

    boolean which if set equal to true enables to monitor the relative change of the estimates of lambda Gamma and Omega in each iteration. The default value of pa.msg is false, that is nothing is displayed in each iteration.
    k

    the number of groups.
    v

    the number of variables.
    userepmat

    scalar, which specifies whether to use implicit expansion or bsxfun. pa.userepmat =2 implies implicit expansion, pa.userepmat=1 implies use of bsxfun. The default is to use implicit expansion (faster) if verLessThanFS('9.1') is false and bsxfun if MATLAB is older than 2016b.

    Data Types: struct

    nocheck — check in input option pa. Boolean.

    Specify whether it is necessary to check the input fields in previous input option pa. If nocheck is false (default is true) no check is performed on input structure pa.

    Data Types: Boolean

    Optional Arguments

    lmd — determinants. Vector.

    Initial estimates of (constrained) determinants

    Example: [ 2 4 6]

    Data Types: double

    OMG — rotation matrices. p-by-p-by-k array.

    p-by-p-by-k array containing the preliminary estimates of the rotation matrices for the k groups. If common rotation is imposed (third letter is equal to E), OMG(:,:,1)=...=OMG(:,:,k).

    Example: .5*hadamard(4)

    Data Types: double

    Output Arguments

    expand all

    Sigma —constrained covariance matrices. p-by-p-by-k array

    p-by-p-by-k array containing the k covariance matrices for the k groups. See section 'More About' for the notation for the eigen-decomposition of the component covariance matrices.

    lmd —restricted determinants. Vector

    Row vector of length $k$ containing restricted determinants.

    More precisely, the $j$-th element of lmd contains $\lambda_j^{1/p}$. The elements of lmd satisfy the constraint pa.cdet in the sense that $\max(lmd) / \min(lmd) \leq pa.cdet^{(1/p)}$. In other words, the ratio between the largest and the smallest determinant is not greater than pa.cdet. All the elements of vector lmd are equal if modeltype is E** or if pa.cdet=1.

    OMG —constrained rotation matrices. p-by-p-by-k array

    p-by-p-by-k array containing the k rotation matrices for the k groups. If common rotation is imposed (third letter is equal to E), OMG(:,:,1)=...=OMG(:,:,k).

    GAM —constrained shape matrix. 2D array

    Matrix of size p-by-k containing in column j the elements on the main diagonal of shape matrix $\Gamma_j$. The elements of GAM satisfy the following constraints: The product of the elements of each column is equal to 1.

    The ratio among the largest elements of each column is not greater than pa.shb.

    The ratio among the second largest elements of each column is not greater than pa.shb.

    ....

    The ratio among the smallest elements of each column is not greater than pa.shb.

    The ratio of the elements of each column is not greater than pa.shw.

    More About

    expand all

    Additional Details

    The notation for the eigen-decomposition of the component covariance matrices is as follows \[ \Sigma_j= \lambda_j^{1/p} \Omega_j \Gamma_j \Omega_j' \qquad j=1, 2, \ldots, k \]

    The dimension of matrices $\Omega_j$ (rotation) and $\Gamma_j$ (shape) is $p\times p$.

    $c_{det}=$ scalar, constraint associated with the determinants.

    $c_{shw}=$ scalar, constraint inside each group of the shape matrix.

    $c_{shb}=$ scalar, constraint among groups of the shape matrix.

    Note that if you impose equal volumes $c_{det}=1$. Similarly, if you impose a spherical shape $c_{shw}=1$.

    We also denote with

    [1] $\Sigma_B$ the 3D array of size $p\times p \times k$ containing the empirical covariance matrices of the $k$ groups, before applying the constraints coming from the 14 parametrizations. In the code $\Sigma_B$ is called $SigmaB$. The $j$-th slice of this 3D array of size $p\times p$ is denoted with symbol $\hat \Sigma_j$.

    [2] $\Omega$ the 3D array of size $p\times p \times k$ containing the rotation matrices of the $k$ groups. In the code $\Omega$ is called $OMG$. The $j$-th slice of this 3D array of size $p\times p$ is called $\hat \Omega_j$.

    [3] $\Gamma$ the $p\times k$ matrix containing in column $j$, with $j=1, 2, \ldots, k$, the diagonal elements of matrix $\Gamma_j$ (shape matrix of group j). In the code matrix $\Gamma$ is called GAM.

    After the application of this routine, the product of the elements of each column of matrix GAM is equal to 1.

    The ratio among the largest (second largest, ...smallest) elements of each column is not greater than $c_{shb}$ (pa.shb).

    The ratio of the elements of each column is not greater than $c_{shw}$ (pa.shb). All the columns of matrix GAM are equal if the second letter of modeltype is E. All the columns of matrix GAM are equal to 1 if the second letter of modeltype is I.

    [4] niini the vector of length $k$ containing the number of units (weights) associated to each group.

    [5] $\lambda$ = the vector of length $p$ containing in the $j$-th position $\lambda_j^{1/p}=|\Sigma_j|^{1/p}$. In the code vector $\lambda$ is called $lmd$.

    The elements of lmd satisfy the constraint pa.cdet in the sense that $\max(lmd) / \min(lmd) \leq pa.cdet^{(1/p)}$. In other words, the ratio between the largest and the smallest determinant is not greater than pa.cdet. All the elements of vector lmd are equal if modeltype is E** or if $c_{det}=1$ (pa.cdet=1).

    References

    Garcia-Escudero L.A., Mayo-Iscar, A. and Riani M. (2020). Model-based clustering with determinant-and-shape constraint, Statistics and Computing, vol. 30, pp. 1363–1380, https://link.springer.com/article/10.1007/s11222-020-09950-w

    Garcia-Escudero L.A., Mayo-Iscar, A. and Riani M. (2022). Constrained parsimonious model-based clustering, Statistics and Computing, vol. 32, https://doi.org/10.1007/s11222-021-10061-3

    See Also

    | | |

    This page has been automatically generated by our routine publishFS