RE: another vectorization challenge

[Tim Rueth]

> -----Original Message-----
> From: Jaroslav Hajek [mailto:address@hidden 
> Sent: Friday, August 27, 2010 1:41 AM
> To: address@hidden
> Cc: address@hidden
> Subject: Re: another vectorization challenge
> 
> On Fri, Aug 27, 2010 at 3:36 AM, Tim Rueth <address@hidden> wrote:
> > For those of you who like to solve vectorization problems, 
> I'm stuck 
> > on the
> > following:
> >
> > Let's say I have a vector with values that range from -10 to 10:
> >
> > cpg_v = 20*rand(1,n) - 10;
> >
> > Now, I want to create another vector cr_v of the same size with the 
> > following logic:
> >
> > Whenever cpg_v drops below "cr_thd," then cr_v should get a 
> 1.  Okay, 
> > that's
> > easy:
> >
> > cr_v = zeros(1,length(cpg));
> > cr_v (cpg_v < cr_thd) = 1;
> >
> > I also want to know when cpg_v goes back above "rec_thd."  
> That's easy, too:
> >
> > rec_v (cpg_v > rec_thd) = 1;
> >
> > But here's the tricky part:  Reading cpg_v from 1:end, once 
> a "1" is 
> > encountered at the corresponding location in cr_v, then 
> cr_v should be 
> > assigned a "2" after that location (unless cpg_v drops below cr_thd
> > again) until the point where cpg_v is greater than "rec_thd."  Note 
> > that a "2" shouldn't be assigned until there's a 1 in cr_v, and the 
> > run of 2's should stop once a 1 is encountered in rec_v.  Here's a 
> > simple example (with cr_thd = -4, rec_thd = 4), and the 
> desired output:
> >
> 
> I usually like to help, but I don't understand your tricky 
> part description. If I did, 4th column should end with 2's.
> 
> >     cpg_v        cr_v        rec_v    desired cr_v
> > -------------------------------------------------------------------
> >    6.45824   0.00000   1.00000   0.00000
> >    7.42197   0.00000   1.00000   0.00000
> >   -2.41237   0.00000   0.00000   0.00000
> >    4.85508   0.00000   1.00000   0.00000
> >    3.78177   0.00000   0.00000   0.00000
> >   -3.24793   0.00000   0.00000   0.00000
> >   -5.63435   1.00000   0.00000   1.00000
> >   -2.10278   0.00000   0.00000   2.00000
> >   -0.80376   0.00000   0.00000   2.00000
> >    7.97556   0.00000   1.00000   0.00000
> >   -5.11177   1.00000   0.00000   1.00000
> >   -4.56460   1.00000   0.00000   1.00000
> >   -8.43245   1.00000   0.00000   1.00000
> >    4.85563   0.00000   1.00000   0.00000
> >   -2.17610   0.00000   0.00000   0.00000
> >    4.87113   0.00000   1.00000   0.00000
> >    6.78267   0.00000   1.00000   0.00000
> >    9.30517   0.00000   1.00000   0.00000
> >    5.31377   0.00000   1.00000   0.00000
> >   -4.93238   1.00000   0.00000   1.00000
> >    2.10757   0.00000   0.00000   2.00000
> >    6.94569   0.00000   1.00000   0.00000
> >    1.67053   0.00000   0.00000   0.00000
> >   -7.23665   1.00000   0.00000   1.00000
> >   -5.43656   1.00000   0.00000   1.00000
> >   -9.79481   1.00000   0.00000   1.00000
> >    0.45726   0.00000   0.00000   2.00000
> >   -2.14684   0.00000   0.00000   1.00000
> >   -0.75054   0.00000   0.00000   1.00000
> >   -2.59986   0.00000   0.00000   1.00000 Obviously, this is 
> easy to do 
> > in a for-loop, but is there a way to do it just with vectors?
> >
> 
> If you can do it with a for loop, why don't you post it here? Not only
> that can clarify your intent, but can also serve as an alternative
> code to check against.
> 
> regards
> 
> -- 
> RNDr. Jaroslav Hajek, PhD
> computing expert & GNU Octave developer
> Aeronautical Research and Test Institute (VZLU)
> Prague, Czech Republic
> url: www.highegg.matfyz.cz

Oh, yes, sorry for any confusion.  This should make it easier to understand:

Here's a for-loop that I wrote, followed by another test scenario.  Even
though I show cr_thd and rec_thd as scalars, they are actually matrices
produced from ndgrid(), so can you not only vectorize this for-loop, but
also ensure that it will work when cr_thd and rec_thd are n-dim arrays?
That would be awesome!

# create a test scenario
n = 30;                         # number of elements to test
cr_thd = -4;                    # crash threshold
rec_thd = 4;                    # recovery threshold
cpg = 20*rand(n,1) - 10;

# initialize the array
cr_v = zeros(n,1);
cr_v(1) = cpg(1) < cr_thd;      # specify first element

# loop through the rest of the elements
for i = 2:n     
        if ( (cr_v(i-1) == 1) | (cr_v(i-1) == 2) ) & (cpg(i) <= rec_thd)
                cr_v(i) = 2;    # have not recovered yet
        endif;
        
        if cpg(i) < cr_thd
                cr_v(i) = 1;    # crashed again (which can happen again even
if not recovered)
        endif;
endfor;

fid = fopen("temp.txt",'wt');
fdisp(fid,sprintf("   cpg_v\tcr_v"));
matrix = [cpg cr_v];
fdisp(fid,matrix);
fclose(fid);

   cpg_v        cr_v
   5.27733   0.00000
  -3.72553   0.00000
   0.21305   0.00000
   5.21520   0.00000
  -2.73123   0.00000
  -4.15256   1.00000
  -9.61998   1.00000
   7.03937   0.00000
   2.59400   0.00000
   4.20354   0.00000
  -3.62042   0.00000
   5.50072   0.00000
   5.77591   0.00000
   5.19524   0.00000
   1.43837   0.00000
   3.60592   0.00000
  -6.17330   1.00000
  -6.33281   1.00000
   0.86042   2.00000
  -4.83423   1.00000
  -2.84996   2.00000
  -2.06245   2.00000
   0.64466   2.00000
  -0.73008   2.00000
   1.53458   2.00000
   8.15784   0.00000
  -4.50903   1.00000
   5.88615   0.00000
   5.99068   0.00000
   1.15822   0.00000

Thanks,

--Tim