| Re: Help optimizing the nested for loops [message #42761] |
Fri, 25 February 2005 07:09  |
Craig Markwardt
Messages: 1869
Registered: November 1996
|
Senior Member |
|
|
Messon Gbah <gbah@umich.edu> writes:
> Could some one help get rid of nested for loops in the following statements?
> The original code was written in C and I'm trying to port it to IDL.
>
> indx = indgen(n)
> indx[0] = 3
> index[3] = 0
> array = dblarr(n,n)
> alpha = dblarr(n,n)
> beta = dblarr(n)
>
> ;Loop 1
> trace = double(0.0)
> for j=0, n-1 do begin
> vj = indx[j]
> trace += alpha[vj,vj]
> alpha[vj,vj] += flamda ;flamda = a constant
> for k=0,j do alpha[vj,indx[k]] = alpha[indx[k],vj]
> endfor
You can remove one level of loops,
trace = total(alpha[indx,indx])
alpha[indx,indx] += flamda
for j = 0, n-1 do alpha(indx[j],indx) = alpha(indx,indx[j])
If you can convert to unpermuted matrices, then you can use the
TRANSPOSE function,
alpha_prime = (alpha[indx,*])[*,indx]
alpha_prime_transpose = transpose(alpha_prime)
> ;Loop 2
> for j=0, n-1 do begin
> vj = indx[j]
> for k=0,n-1 do array[indx[k],j] =
> alpha[indx[k],vj]/sqrt(alpha[vj,vj]*alpha[indx[k],indx[k]])
> endfor
Again, it's probably worth converting to unpermuted matrices.
alpha_prime_diag = alpha[indx]
for j = 0, n-1 do $
array_prime[*,j] = alpha_prime[*,j] / sqrt(alpha[j,j]*alpha_prime_diag)
> ;Loop 3
> for j=0, n-1 do begin
> vj = indx[j]
> b[vj] = T[vj] ;T is some init value
> for k=0, n-1 do begin
> vk = indx[k]
> b[vj] += beta[vk]*array[k,j]/sqrt(alpha[vj,vj]*alpha[vk,vk])
> endfor
> endfor
Again, converting to unpermuted matrices,
for j = 0, n-1 do $
b[j] = T[j] + total(beta*array_prime(*,j)/sqrt(alpha_prime_diag[j]*alpha_p rime_diag))
...although it's a little confusing which of your matrices are
permuted and which are not, so that may require some tweaking.
Good luck,
Craig
--
------------------------------------------------------------ --------------
Craig B. Markwardt, Ph.D. EMAIL: craigmnet@REMOVEcow.physics.wisc.edu
Astrophysics, IDL, Finance, Derivatives | Remove "net" for better response
------------------------------------------------------------ --------------
|
|
|
|
| Re: Help optimizing the nested for loops [message #42922 is a reply to message #42761] |
Tue, 01 March 2005 13:16  |
Messon Gbah
Messages: 4
Registered: September 2000
|
Junior Member |
|
|
Hi Craig:
Thanks for your reply and suggestions.
V. Best,
Messon
Craig Markwardt wrote:
> Messon Gbah <gbah@umich.edu> writes:
>
>
>
>> Could some one help get rid of nested for loops in the following statements?
>> The original code was written in C and I'm trying to port it to IDL.
>>
>> indx = indgen(n)
>> indx[0] = 3
>> index[3] = 0
>> array = dblarr(n,n)
>> alpha = dblarr(n,n)
>> beta = dblarr(n)
>>
>> ;Loop 1
>> trace = double(0.0)
>> for j=0, n-1 do begin
>> vj = indx[j]
>> trace += alpha[vj,vj]
>> alpha[vj,vj] += flamda ;flamda = a constant
>> for k=0,j do alpha[vj,indx[k]] = alpha[indx[k],vj]
>> endfor
>
>
> You can remove one level of loops,
>
> trace = total(alpha[indx,indx])
> alpha[indx,indx] += flamda
> for j = 0, n-1 do alpha(indx[j],indx) = alpha(indx,indx[j])
>
> If you can convert to unpermuted matrices, then you can use the
> TRANSPOSE function,
> alpha_prime = (alpha[indx,*])[*,indx]
> alpha_prime_transpose = transpose(alpha_prime)
>
>
>
>
>> ;Loop 2
>> for j=0, n-1 do begin
>> vj = indx[j]
>> for k=0,n-1 do array[indx[k],j] =
>> alpha[indx[k],vj]/sqrt(alpha[vj,vj]*alpha[indx[k],indx[k]])
>> endfor
>
>
> Again, it's probably worth converting to unpermuted matrices.
>
> alpha_prime_diag = alpha[indx]
> for j = 0, n-1 do $
> array_prime[*,j] = alpha_prime[*,j] / sqrt(alpha[j,j]*alpha_prime_diag)
>
>
>
>> ;Loop 3
>> for j=0, n-1 do begin
>> vj = indx[j]
>> b[vj] = T[vj] ;T is some init value
>> for k=0, n-1 do begin
>> vk = indx[k]
>> b[vj] += beta[vk]*array[k,j]/sqrt(alpha[vj,vj]*alpha[vk,vk])
>> endfor
>> endfor
>
>
> Again, converting to unpermuted matrices,
>
> for j = 0, n-1 do $
> b[j] = T[j] + total(beta*array_prime(*,j)/sqrt(alpha_prime_diag[j]*alpha_p rime_diag))
>
> ...although it's a little confusing which of your matrices are
> permuted and which are not, so that may require some tweaking.
>
> Good luck,
> Craig
>
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
