[Scilab-Dev] Question about sparse(ij,v,mn) internals

[Stéphane Mottelet]

Hello Samuel,

Le 11/05/2015 21:16, Samuel Gougeon a écrit :
> Hello Stéphane,
>
> Le 11/05/2015 15:03, Stéphane Mottelet a écrit :
>> Hello Scilab devs/masters,
>>
>> For performance purposes, i need to implement a function which is 
>> missing to Scilab, allowing to set at once all non-zero terms of a 
>> sparse matrix. Some other languages have this feature, e.g. Julia 
>> allows to write such things.
>>
>> A.nzval=v
>>
>> of course this is interesting when the sparsity pattern does not 
>> change but actual element values do. I noticed that the use of 
>> "sparse" to reconstruct matrices in this context is the main 
>> bottleneck of my code... In Scilab the logical name of such a command 
>> should be "spset" and used like this
>>
>> spset(A,v)
>
> I may miss something: how do you specify the position of terms to be 
> set with the v content? Is A already set, and you just want to replace 
> its nz terms? Is .nzval the "list" of positions (in such a way that we 
> could read A(nzval) .= v , and then what is the equivalent with the 
> spset(A,v)
no, nzval is really the field name such that A.nzval has the value v 
sucht that [ij,v]=spget(A). In Julia, the fields of a sparse matrix are

julia> A.
colptr m       n       nzval   rowval

colptr and rowval give the indices for the classical column compressed 
format. The nzval field is of read-write type so that the non zero 
elements can be updated by "A.nzval=v" if v has the correct size and 
corresponds to correctly ordered (column-wise) unique (i(k),j(k),v(k)) 
pairs.

S.


>   syntax (unless nzval -- that could rather be nzpos -- is implicit)?
>
> Anyway, in a more general way,  i agree that an element-wise 
> assignment operator would be very welcome, as well for lists, structs 
> and cells -- say heterogenous -- containers.
>
> Whether
> [A,B,C, D] = ("hello", %t, (1-%s)^2, list("bye", %pi) )
> sets distributively A = "hello", B=%t, C = (1-%s)^2, D = list("bye", %pi)
> yet,
> L = list(); L(3) = "ok"
> L(:) = ("hello", %t, (1-%s)^2)  // does not work
> A pity!
>
> Regards
> Samuel
>
> _______________________________________________
> dev mailing list
> dev at lists.scilab.org
> http://lists.scilab.org/mailman/listinfo/dev


-- 
Département de Génie Informatique
EA 4297 Transformations Intégrées de la Matière Renouvelable
Université de Technologie de Compiègne -  CS 60319
60203 Compiègne cedex