Hi

This normalisation isn't directly part of the actual L2 regularisation - it is 
just setting the overall scaling of the covariance matrix, so that the effect 
of choice of regularisation parameter is not dependent on the overall scaling 
of the original data.

The scaling is just the RMS of the diagonals of the covariance - this is just a 
somewhat arbitrary choice for how to set the overall scaling of the matrix - 
other options such as just taking the mean would probably be fine too.

Cheers, Steve.




> On 17 Aug 2017, at 01:11, Mary Beth <mb.ne...@gmail.com> wrote:
> 
> Hi pals,
> 
> I have a question about the way partial correlations were estimated for the 
> megatrawl using FSLnets. Going through the 'ridgep' section of 
> nets_netmats.m, it looks like the covariance matrix for each subject is 
> normalized by the square root of the mean of the variances squared - yes, no, 
> maybe so?
> 
> from line 88 of nets_netmats.m:
> 
> grot = cov(<component timeseries>);
> grot = grot/sqrt(mean(diag(cov1).^2));
> 
> I'm just trying to figure out why you square the variances before you take 
> the mean. Can someone give me a quick explanation or point me towards a good 
> reference?
> 
> Thanks in advance for your help!
> mb 
> _______________________________________________
> HCP-Users mailing list
> HCP-Users@humanconnectome.org
> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
> 


---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Head of Analysis,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
st...@fmrib.ox.ac.uk    http://www.fmrib.ox.ac.uk/~steve 
<http://www.fmrib.ox.ac.uk/~steve>
---------------------------------------------------------------------------

Stop the cultural destruction of Tibet <http://smithinks.net/>






_______________________________________________
HCP-Users mailing list
HCP-Users@humanconnectome.org
http://lists.humanconnectome.org/mailman/listinfo/hcp-users

Reply via email to