[Coin-standards] Re: data structures (Leonardo B. Lopes)

[Alan King]

Hi Leo --

I would like also to publicly thank you for all the work you are putting
into this.

As to specifics:  Bob Entriken, Steve E. Wright and I developed the basic
data handler for the IBM OSLSE code.  We implemented the Stochastic MPS
"scenarios" standard as follows: all data is converted to "matrix form"
including row/col bounds and the objective.  Then we store changes as
additive updates to the CORE matrix data.   We had lots of discussions
(some heated!) about this, so I believe this method is about as good as you
are going to get for the "scenarios" format.

There are more efficient data handling possibilities when the stochastic
data is extremely regular, as is often the case in financial models.   In
options pricing, one often encounters extremely regular trees --- like
"ternary tree with 30 time steps and space step 1.26".  Or one may wish to
generate the tree topology from data observations (like options prices).

It seems to me now that there should be a hierarchy of data-management.
For example, the specification "ternary tree with 30 time steps and space
step 1.26" could be useful at a high level in the problem description ---
say, to organize the computation between many parallel nodes --- but as we
get closer to actual linear programming, it seems that one would want to
use the S-MPS style of data-management.

Should we make a list of the tree management issues that folks are
encountering and see if there are patterns that can be generalized?

Alan

Alan King
Mathematical Sciences Department
IBM Thomas J. Watson Research Center
(914) 945-1236, 8-862-1236