[pymvpa] saving searchlight map in 0.4.4

[Yaroslav Halchenko]

On Tue, 02 Feb 2010, Sergey.V.Fogelson at Dartmouth.edu wrote:
> system monitor on my ubuntu desktop and the python process's status
> was Uninterruptible and the amount of ram it was using was a
> whopping 10.1 Gigs!
I guess something funny with "unmasking"

> Any idea what may be causing this? Again, my earlier pymvpa version
> (0.4.2) did not have this problem.
damn... surprises of new release :-/ 

> >s1_map = s1(dataset)

wow... my bad... can't believe that we/I managed to miss this issue...
morning might reveal how that became possible (if I am correct) but just
instead of

orig_s1_map = dataset.mapReverse(N.array(s1_map))

do

orig_s1_map = dataset.mapReverse(s1_map.T)

1. array() isn't necessary here I think (should check either in general
and fix example) and it would cause memory duplication
2. .T is critical here -- in 0.4.4 we have got a spurious dimension
added... so to make your code compatible with both I actually suggest
using

orig_sl_map = dataset.mapReverse(N.asanyarray(sl_map).squeeze())

that will just make sure it is some array (and wouldn't duplicate if it
is already) and then would remove that evil bogus dimension.

sorry about trouble -- it is never enough of testing :-/

-- 
                                  .-.
=------------------------------   /v\  ----------------------------=
Keep in touch                    // \\     (yoh@|www.)onerussian.com
Yaroslav Halchenko              /(   )\               ICQ#: 60653192
                   Linux User    ^^-^^    [175555]