Re: [Gretl-users] on gretl's capacity to handle a large matrix
On Tue, 24 Dec 2013, artur.bala.tn wrote:
Thank you. Indeed, gretl only crashed when adding the --cluster=unit
option.
Best,
Artur
<hansl>
list xvars = const PROFI(-1) TAILLE(-1) MTB(-1) VOLAT(-1) D55 D55PROFI
genr unitdum
ols DFMN xvars du_* # It works
ols DFMN xvars du_* --cluster=unit It doesn't
I wasn't able to produce an actual crash with your example (i.e. a
segmentation fault), and valgrind (memory checker) says the code is OK,
but... it did take an inordinate time to calculate the clustered standard
errors (gretl became unresponsive for several minutes).
This is now fixed in CVS and snapshots. It now takes less than 2 seconds
on my laptop to calculate the 602 x 602 clustered variance matrix.
(Admittedly that's with threaded BLAS running 4 threads.)
Allin