search

ecamenen / rgcca-conda

0 stars 0 forks source link

Question: how to deal with missing values #1

Open mbrunin opened 4 years ago

mbrunin commented 4 years ago

Hello,

When you perform the deflation of a block in order to compute its second component, you multiply the matrix of the orthogonal projection on the orthogonal of the first component and this block.

How do you compute this multiplication of two matrices if this block contains missing values ?

Best regards Maxime BRUNIN

ecamenen commented 4 years ago

Hi, Please find below the response of the person working on this missing values issue.

If you are interested in working with our tool and you are familiar with conda you can pull the last release branch or, preferably, directly the latest conda version on this specific channel (last master build: noarch/r-rgcca-2.3.0-r35_010.tar.bz2) rather than the one on conda-forge. For a wider use of the RGCCA features under R (including our new missing values features), an increment of the current CRAN version of the RGCCA package is under development. A relatively stable version can be the one tagged "2.3.0". However, I strongly advise you to wait for the end of this new release (~1 month) or the creation of a more adequate release branch and its associated documentation. We can stay in touch and keep you informed of our progress if you wish. @ecamenen (etienne.camenen@icm-institute.org)

Hello Maxime, Thank you very much for your interest in our methods. If the matrix contains missing values, they are not taken into account in the projection. As in NIPALS approaches, the regressions are computed on available data only. That is equivalent to replace missing values by zero in the centered matrices. Other approaches such as imputations could also be used. We are currently writing a paper about these methods and we would be happy to send it to you (when finished) if you are interested. Do not hesitate if any other questions. Best regards, @cpeltier (caroline.peltier@icm-institute.org)