Re: [Icatb-discuss] ICASSO parameters

[Srinivas R. <sra...@mr...>]

Hi Sam,

Both Infomax and Fast ICA determine super gaussian sources. You could use
either one of the algorithms. In older versions of GIFT, centrotype
estimates were used. Min and max cluster parameters are used along with the
stability index to determine best stable run. We use best stable run
estimates instead of centrotype. There are some issues with using
centrotype estimates when using higher model order. Please see
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3222740/ for more info.

Thanks,
Srinivas

On Tue, Nov 11, 2014 at 6:58 AM, Sam Evans <sam...@gm...>
wrote:

> Dear gift list members
>
> I have conducted an ICA analysis which identifies 34 components using the
> MDL criterion.
>
> I now want to test the stability of these networks using ICASSO.  I have
> two questions:
>
> (1) should I use fastICA or Infomax as the ICA algorithm?  Is there a
> particular reason to use either?
>
> (2) For the parameters of ICASSO, I was planning to use:
> - random iterations
> - 50 iterations
> - min cluster = 0.8.*50 (i.e. 0.8* number of iterations, based on a
> previous post to the forum) = 40
> - max cluster = 0.8.*50=40
>
> Are these parameters appropriate - I note that most studies do not report
> the cluster parameters...
>
> Thanks for taking the time to read my questions.
>
> Sam Evans
>
> ------------------------------------------------------------------------------
> Comprehensive Server Monitoring with Site24x7.
> Monitor 10 servers for $9/Month.
> Get alerted through email, SMS, voice calls or mobile push notifications.
> Take corrective actions from your mobile device.
>
> http://pubads.g.doubleclick.net/gampad/clk?id=154624111&iu=/4140/ostg.clktrk
> _______________________________________________
> Icatb-discuss mailing list
> Ica...@li...
> https://lists.sourceforge.net/lists/listinfo/icatb-discuss
>