Thanks for that. I took it, and played around with it a bit to get it
to work. [Errors I found were: x and y don't concatinate in the line
'array=transpose([[x],[y]])' and I found I had to comment away the
/ISOTROPIC in the plotting.) Unfortunately, it seems that cluster
seperates on a purely 1 dimensional basis. I tried discarding the
histogram related code in favor of a much simpler system in case that
was the problem, and it still didn't work. If you look at the data set
I provided, the problem should be self evident.
Incidentally, I replaced everything from
h=histogram(c,reverse_indices=ri) down to the second to last line with:
--
plot,x,y,psym=2
bmax=max(array[0,*],maxsubsc)
goodc=c[maxsubsc]
keep=where(c[*] eq goodc)
--
I feel that my code may be a tad more efficient, though I don't know
how efficient the WHERE command is.
Anywho, I'm looking CLUSTER_TREE right now, which shows some more
promise. If I understand it correctly, it works using distance appart,
not coordinates which is a bit more useful, I think, for my problem.
I'm just not sure how I can take the output and turn it into a set of
clusters.
Thanks for all the help!
- Rob
JD Smith wrote:
> On Tue, 08 Aug 2006 15:20:13 -0700, rdellsy wrote:
>
>> I'm a tad confused about what you're suggesting. I'll try and work it
>> out, but I'm still fairly new to IDL, so if you could give an IDL or
>> pseudo-code example of what you're trying to explain, I would
>> appreciate. If that's too much work, I understand, and I'll just try to
>> puzzle it out on my own.
>
> You might find much of what you need in the HISTOGRAM tutorial:
>
> http://www.dfanning.com/tips/histogram_tutorial.html
>
> But before you go that route, you might first try the CLUSTER function
> in IDL (which I just read up on). Here's an example using a fake
> clustered data set with 5 clusters. You'll probably have to experiment
> with the number of clusters.
>
> JD
>
> tvlct,[255,0,0,0,255,255],[0,255,0,255,255,0],[0,0,255,255,0 ,255],1
> n_clust=5
>
> ;; Make some flake clustered data
> if n_elements(x) ne 0 then begin
> n=1000
> clust_fwhm=.2
> cposx=randomu(sd,n_clust) & cposy=randomu(sd,n_clust)
> cind=fix(randomu(sd,n)*n_clust)
>
> x=clust_fwhm
> fac=2*sqrt(2*alog(2))
> x=randomn(sd,n)*clust_fwhm/fac+cposx[cind]
> y=randomn(sd,n)*clust_fwhm/fac+cposy[cind]
> endif
>
> array=transpose([[x],[y]])
> w=clust_wts(array,N_CLUSTERS=n_clust)
> c=cluster(array,w)
> h=histogram(c,REVERSE_INDICES=ri)
> nh=n_elements(h)
>
> plot,x,y,PSYM=4,/ISOTROPIC
>
> cen=make_array(2,nh,VALUE=!VALUES.F_NAN)
> for i=0,nh-1 do begin
> if ri[i+1] eq ri[i] then continue
> take=ri[ri[i]:ri[i+1]-1]
> oplot,x[take],y[take],PSYM=4,COLOR=i+1
> cen[0,i]=[mean(x[take]),mean(y[take])]
> endfor
>
> ;; Find the lower right cluster
> void=max(cen[0,*]-cen[1,*],lrc,/NAN)
>
> ;; Highlight it
> keep=ri[ri[lrc]:ri[lrc+1]-1]
> oplot,x[keep],y[keep],PSYM=6,SYMSIZE=2
>
> END
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
