[Help-glpk] Processing glpsol output with R (was: Need help on interval planning constraint)

[Robbie Morrison]

Hello Jeffrey

The quick answer is "not hard at all".

RECOVERING JUST THE VALUES

I suggest starting with the 'glpsol' option:

  --write filename : write solution in plain text format

The first thing to determine is whether the problem is
pure-LP or MIP.  If not already known, this can be
undertaken by counting the number of values in line
two: three values indicates a pure-LP and two indicates
an MIP (whether this is robust in the event of the
solver returning failure or similar needs confirming).

The next step is to parse the entire file according to
the formatting rules for the underlying API call:

  pure-LP problem : 'glp_write_sol'
  MIP problem     : 'glp_write_mip'

Consult the API manual for the actual details here.

Parsing should be straightforward using double type
arrays and matrices from R, in conjunction with the
m x n information previously captured from line one.

If one wants the LP-relaxation solution for an MIP
problem, then I guess one would need to subsequently
deploy '--nomip' and repeat the parsing task.

[As a suggestion to Andrew, it might be cleaner for the
'--write' option to state something like "LP" or "MIP"
in the opening line to unambiguously indicate the
problem class -- or perhaps even give a finer
resolution, for instance "mixed-integer", "mixed-01",
etc).  Note too that the now depreciated 'lpx_get_class'
call used to provide at least some of this information.]

The R command that could be used to read in the raw
data is 'readLines()':

  data  <- readLines(filename)  # default is to read all
  lines <- length(data)
  message("lines read     : ", lines)
  if ( lines == 0 ) warning("no lines read")

Also useful might be 'strsplit()', 'unlist()', and
'as.numeric()' for string to number coercion.

OBTAINING MORE INFORMATION

It would also be quite possible to parse the results
from the existing 'glpsol' option:

  --output filename : write solution in printable format

Parsing this data would take more work but would also
reveal more information, such as row and col labels
(names).

The current 'glpsol' practice of wrapping the output at
text column 65 also makes the parsing task a little
more awkward (as occurs in the presence of long row and
col labels) but not impossible.

[Perhaps a new 'glpsol' option might help here:

  --meta filename : write problem metadata in parsable
                    key/value format

This option would output the problem name, row and col
labels and so-forth for use in conjunction with option
'--output'.  It could equally apply to unsolved and
solved problem instances, but with the current solution
status in the latter case.]

READING THE PROBLEM

The flip question (although not stated) is that of
recovering the original problem using R via an exported
MPS, CPLEX, or (not currently documented and/or fully
implemented) OPB format.

R itself already supports LP importation as part of
either its core distribution or via contributed code.
Here are some packages although I have no idea as to
their respective statuses, except I think 'Rglpk' has
yet to catch up with the new APIs:

  package 'Rcplex' - R interface to the CPLEX
    optimization library

  package 'linprog' - linear programming / optimization

  package 'Rglpk' - high level R interface to the MPS
    and CPLEX_LP reader of the GNU Linear Programming
    Kit (GLPK).

CLOSURE

I don't use 'glpsol', but I imagine that some modeling
tasks are adaptive and that this adaptation might more
easily take place in an external scripting environment
than in MathProg.  Hence, script-friendly status and
results recovery could be of benefit here.

with best wishes to all
Robbie

ps: Andrew: the "glp_write_sol" on page 99 of the 4.41
    API manual should read "glp_write_mip".

------------------------------------------------------------
To:          Robbie Morrison <address@hidden>
Subject:     Re: [Help-glpk] Need help on interval planning constraint
From:        Jeffrey Kantor <address@hidden>
Date:        Tue, 22 Dec 2009 16:58:32 -0500
------------------------------------------------------------

Robbie,

How hard would it be to write a standard import script
for R?  That is, something to take glpsol's standard
report and digest it into something that could be
directly manipulated in R.

On Tue, Dec 22, 2009 at 3:51 PM, Robbie Morrison
<address@hidden>wrote:

> >
> > Hello GLPK users
> >
> > A plug for R : I use R for visualization and like it very much.  The
> > environment is comparable with Matlab, but the language is nicer.  I
> > either run it from a shell or from within Emacs (recommended only if
> > you know and love this editor) using the ess-mode.
> >
> > with best wishes, Robbie

---
Robbie Morrison
PhD student -- policy-oriented energy system simulation
Technical University of Berlin (TU-Berlin), Germany
University email (redirected) : address@hidden
Webmail (preferred)           : address@hidden
[from IMAP client]