incubator-ooo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pedro Giffuni <...@apache.org>
Subject Re: [code] 4.x Feature planning: MTL proposal
Date Tue, 31 Jan 2012 15:53:09 GMT
--- Mar 31/1/12, Rob Weir <robweir@apache.org> ha scritto:

> 
> How would this be exposed in Calc?  Are you thinking of
> having it handle the matrix spreadsheet functions like
> MMULT() and MINVERSE()?

Yes at first it would replace the existing matrix
operations. In a spreadsheet everything is composed
of matrices so even managing the existing calculations
with MTL would bring some level of simplification and
code reduction

> Or were you thinking of exposing some more advanced
> functions or tools, like eigenvectors/eigenvalues
> calculations?
>

That would be a latter step: arguably this would be
of great use for a small group of people. An accountant,
for example, will have no use for it.
 
> It would be good to understand the advantages of changes
> the code here.  I assume it would be for better performance
> with large spreadsheets, especially now that OOo supports
> 1 million rows of data.
> 

Normally, doing code that scales well with massive amount
of data actually slows down small size calculations but
small size calculations are already so fast to compute that
no one will notice. The difference will be in big problems,
let's say we save an half an hour in a calculation that takes
28 hours.

Such calculations can be very common in scientific
environments and in niche markets like the oil industry:
there, people have specialized software that they just fork
and forget about. The advantage of Calc is the interface,
and while people don't appreciate having a spreadsheet open
for 28 hours, no matter the size, this will be very useful
for medium sized calculations and maybe to interface with
specialized modules (volunteer can look at MUMPS).

To make some idea, I think this would compete to some extent
with Matlab-like toolsets, where people have a general matrix
solving environment and additional toolsets available for
specialized problems.

Pedro.

Mime
View raw message