Re: [ferret_users] eof_stat error message

[Ansley Manke <ansley.b.manke@xxxxxxxx>]

Hi Domingos,
As Fabian says, this probably is a lack of memory. The error message 
listing a negative number of words of memory indicates an overflow in 
the value that was computed.  (What version of Ferret are you running? 
Some of these error message have I believe been improved in the most 
recent versions.).  The EOF functions use a lot of resources, because 
they operate on a matrix that is NX * NY *NT square

The Ferret command SET MEMORY/SIZ= is allocating whatever memory is 
available at the moment you issue the command. Say SHOW MEMORY 
afterwards to see what memory you were actually able to allocate.

With any function, it's often a good idea to do a small test case first 
to see how it works.  Try subsetting your data, perhaps something like this:

  yes? LET var_subset = var[I=1:256:10,J=1:192:10,L=1:240:12] 
  yes? LET eof_stat_subset = EOF_STAT(var_subset)

to do a much smaller problem, and explore the results.

Fabian Lienert wrote:
> Hi Domingos,
>
> I don't see how you used eof_stat, but the error message tells
> you that there is not enough RAM for the computation (RAM, not swap).
>
> If the resolution of the grid of the input data is high, inverting 
> such a matrix in order to find the EOFs uses a lot of RAM.
> Calculating the EOFs in the northern Pacific (20N,65N) for 330 time 
> steps needs about 800 MB of RAM
> if the data is on a T63 grid (lon = 256, lat = 192).
> Higher resolved input data leads to the use of more RAM. This behavior 
> is not linear.
>
> Fabian
>
>
> Domingos Urbano wrote:
> > Hello Ferrets,
> >
> > I have a time series of 240 time steps, 40N to 40S, global but z@ave, 
> > and want to compute EOF from pacific and atlantic region between 
> > 20N,20S region.
> >
> > But I had problem with the first step!
> > What am I doing wrong?
> > Is my external functions not installed correctly or my data is too 
> > big? I already set memory/size=1000 but nothing changed.
> > Any help is welcome.
> >
> > Cheers,
> > Domingos.
> >
> >
> > yes? list/i=1/j=1 estat
> >
> > ERROR in efcn_compute() allocating -443066952 bytes of memory
> >       work array 4:  X=1:2036716910, Y=1:1, Z=1:1, T=1:1
> >  **ERROR: error in external function
> > yes?
> >
> >
> > --
> > ---------------------------------------------------------------
> > Dr. Domingos Urbano
> > Grupo de Modelagem Acoplada Oceano-Atmosfera - GMA
> > Centro de Previsão de Tempo e Estudos Climáticos - CPTEC
> > Instituto Nacional de Pesquisas Espaciais - INPE
> >
> > Rodovia Presidente Dutra, km 40
> > 12630-000 - Cachoeira Paulista - SP - Brasil
> > tel: 55(12)3186.8552 fax: 55(12)3101.2835
> > www.cptec.inpe.br <http://www.cptec.inpe.br>   www.inpe.br/ingles/ 
> > <http://www.inpe.br/ingles/>
> > ------------------------------------------------------