Fwd: Three Question about PCA by using ViSta 7.9.2.5
Pedro Valero
From: Hisham Abu Samra <hisham....@gmail.com>
Date: 2012/1/25
Subject: Three Question about PCA by using ViSta 7.9.2.5
To: Pedro.Va...@uv.es
Dear Prof. Valero-Mora,
I would like to take this opportunity to thank you and your team very much for all the efforts put into ViSta and making it available to researchers free of charge.
I am doctoral student in the University of Siegen in Germany and have recently discovered the beauty of Multivariate Analysis and started using ViSta to carry out Principal Components Analysis (PCA) on spectroscopy generated data. I am writing to kindly ask you three questions regarding the implementation of PCA and dataset treatment using the latest online version: ViSta 7.9.2.5.
Q1: My data consists of 150 observations (samples with different chemical compositions) and 130 variables (spectroscopic chemical composition-related signal intensities).
The dataset was first pre-treated in such a way that the signal intensities are normalised for each sample. Importing the data to ViSta and carrying out PCA was successful.
However, PCA analysis failed on the same data after mean-centering (that is each single variable is subtracted of its mean value across all samples). Surely there are no issues with importing the data. The following error messages appeared:
(Error: Floating point exception. Happened in: #<Subr-LINPACK-DSVDC: #1d5ff88>)
On the other hand, this problem does not seem to occur when the input data matrix size is reduced to 150 Obs. X 110 Var. Could this be related to the input matrix size which ViSta could process?
I am not a developer and don’t have experience with programming, would it be possible to provide me with some feedback as to why this occurred and how I could overcome this issue?
Q2: In certain cases , I face the problem of having more variables than observables.
Is there a plausible way with which I could still implement PCA without having to reduce the matrix size?
Q2: Is it possible to carry out specific data treatment prior to performing PCA using ViSta to obtain for example:
(i) mean-centering of the dataset.
(ii) variance scaling (each single variable is divided by its standard deviation in the input dataset).
Thank you very much in advance for your time and your feedback.
Best regards,
Hisham
Hisham Abu Samra
Email: hisham....@gmail.com