Nway104
所属分类:matlab编程
开发工具:matlab
文件大小:146KB
下载次数:83
上传日期:2006-03-25 09:01:16
上 传 者:
bigsheng
说明: 多维数据处理:MATLAB源程序用于处理多维数据
(multidimensional data processing : MATLAB source for dealing with multidimensional data)
文件列表:
Nway104\unimodalcrossproducts.m (4543, 1999-11-15)
Nway104\chkpfdim.m (771, 1998-07-21)
Nway104\ckron.m (530, 1998-08-26)
Nway104\cmatrep.m (634, 1999-11-15)
Nway104\complpol.m (143, 1999-10-11)
Nway104\contents.m (2459, 1999-10-11)
Nway104\corcond.m (5919, 1999-10-11)
Nway104\coredian.m (397, 1998-08-26)
Nway104\coreswdn.m (839, 1998-08-26)
Nway104\corevarn.m (419, 1998-08-26)
Nway104\derdia3.m (1820, 1998-08-26)
Nway104\derswd3.m (2135, 1998-08-26)
Nway104\dervar3.m (3077, 1998-08-26)
Nway104\dtld.m (3321, 1999-10-11)
Nway104\explcore.m (1703, 1999-11-15)
Nway104\fac2let.m (2452, 1999-01-22)
Nway104\fastnnls.m (1908, 1999-10-11)
Nway104\fnipals.m (6491, 1999-01-22)
Nway104\getindxn.m (480, 1999-10-11)
Nway104\gram.m (2443, 1999-10-11)
Nway104\gsm.m (971, 1998-08-26)
Nway104\ini.m (7336, 2000-09-11)
Nway104\inituck.m (5854, 1999-01-22)
Nway104\maxdia3.m (4200, 1998-08-26)
Nway104\maxswd3.m (3811, 1998-08-26)
Nway104\maxvar3.m (4191, 1998-08-26)
Nway104\meannan.m (601, 1999-10-11)
Nway104\missmean.m (974, 1999-10-11)
Nway104\missmult.m (873, 1999-10-11)
Nway104\misssum.m (1156, 1999-10-11)
Nway104\monreg.m (2365, 1999-11-15)
Nway104\ncosine.m (2383, 1998-09-04)
Nway104\neye.m (1478, 1999-01-22)
Nway104\nident.m (474, 1998-08-26)
Nway104\nmodel.m (5284, 1999-10-11)
Nway104\nonneg.m (2416, 1999-01-22)
Nway104\noptiopf.mat (76, 1998-08-26)
Nway104\noptiot3.mat (108, 1998-08-26)
Nway104\normit.m (545, 1999-10-11)
Nway104\npls.m (7720, 2000-09-11)
... ...
IMPORTANT NOTES ON THE N-WAY TOOLBOX ver. 1.03
There are some important details that are necessary to know in order
to be able to use the N-way toolbox properly. These are given here - please read
carefully before using the toolbox.
CONDITIONS
The toolbox is freeware and may be utilized (but not modified) if proper reference
is given to the authors.
The N-way Toolbox for MATLAB ver. 1.03, http://www.models.kvl.dk/
R. Bro & C. A. Andersson
Food Technology
The Royal Veterinary and Agircultural University
DK-1958 Frederiksberg
Denmark
WARRANTY
In short, no guarantees, whatsoever, are given for the quality
of this toolbox or for the consequences of its use. It is inevitable
that there will be some bugs, but we have tried to test the
algorithms thoroughly.
WHERE DOES THE TOOLBOX WORK?
The toolbox has been tested on MATLAB 4.2 and 5. in Windows95 only. No
use has been made of the arrays supported by MATLAB 5, since this would prevent
the use of the algorithms on earlier versions of MATLAB. However, next major
update of the toolbox is likely to utilize arrays (unless our users tell us not
to).
SETTING UP THE TOOLBOX
In order to install the toolbox copy the main files to a directory (e.g. NWAY)
and copy the additional auxiliary files into a subdirectory called PRIVATE.
Make sure that both these paths are included in MATLAB's path (in MATLAB 5 the
'private' path is automatically included)
DATA INPUT
In order to input the N-way data into the algorithm they must be arranged
as a matrix. This is described in several places. Consider for example
a situation where you have measured several samples on a spectrometer.
This is an ordinary two-way data set which can be stored in a matrix,
X1. Suppose the same data are also measured similarly but at two
other pH values. Two more data set are then available: X2 and X3.
Together these three matrices can be arranged as a three-way array
(sample x spectra x pH), but in MATLAB they will be held as a matrix
X = [X1 X2 X3].
Define DimX = [I J K], that is a three-vector where I is the number
of samples, J the number of wavelenghts, and K the number of pH values
(three in this case). Then X is really an I x J x K array but is held
as an I x JK matrix. To calculate, say, a two-component PARAFAC model
write
Factors = parafac(X,DimX,2);
Note, that the row-mode of the data matrix must correspond to the first
mode of the array, while the column mode of the matrix is holding the
remaining (confounded) modes. Note also how these confounded modes
are arranged. The second mode is the one with its indices 'running fastest'.
MODEL OUTPUT
All algorithms for fitting models use a specific format for the
estimated loadings. This format has been chosen because it enables
a straightforward generalization to N-way arrays.
Consider for a simple example ordinary two-way PCA. A
typical MATLAB format for a NIPLAS-based PCA algorithm could be
[T,P]=fnipals(X,F)
where X is the (I x J) data matrix and F the number of components.
Normally the scores, T, and loadings, P, are given in two separate
matrices. This is a very natural way of outputting the parameters.
For the N-way models all loadings are output in one vector called
'Factors'. The first elements of Factors hold the vectorized first
loading matrix, the following the vectorized second loading matrix,
etc. In case of the PCA model above, Factors would read
Factors = [vecT;vecP] = [T(:);P(:)]
For a three-way PARAFAC model 'Factors' would read
Factors = [A(:);B(:);C(:)]
where A is the first-mode loading matrix etc.
For a three-way Tucker3 model Factors would be given as for the
PARAFAC model, but additionally a core array would be output.
[Factors,G]=tucker(X,DimX,F);
For a tri-PLS2 model (three-way X, two-way Y) two vectors are
defined
Xfactors = [T(:);Wj(:);Wk(:)];
Yfactors = [U(:);Q(:)];
In order to convert the vectorized parameters to ordinary scores and
loading matrices the M-file FAC2LET (factors to letters) is used.
MISSING DATA
For all algorithms the same flag is used for missing elements, namely
NaN. If you have a data set, X, where missing elements are, e.g.,
designated by the number -9999, you can easily modify the data as
X(find(X==-9999))=NaN*find(X==-9999);
SUPPORT
We are VERY interested in and dependent on feedback from the users. If
you have problems running the toolbox please supply screendumps as well as
version number of toolbox, MATLAB, and operating system before contacting
us. We will do the utmost to help overcoming the problems. In the rare event
that the support required is very time-consuming we will have to charge for
this service.
The authors may be contacted by email:
claus@andersson.dk (primarily Tucker and application/helper programs)
rb@kvl.dk (primarily PARAFAC/N-PLS)
近期下载者:
相关文件:
收藏者: