Mac install advice

[Roger Koenker]

Now that Rmosek supports additively separable problems, it is back on
my radar, and my experience
so far is very positive. I've used Mosek for several years via an R
interface with matlab, Rmosek is
a great step forward. I have two bits of advice about installation
and one question:

To get started I needed to do two things: remember that my Mosek
installation was 32 bit so I needed
to invoke R32, rather than my default R <- R64, the second was to
define

MOSEKLM_LICENSE_FILE=/usr/local/mosek/6/licenses/mosek.lic

this is most conveniently done in a .Rprofile file in the directory
that is being used for the project.
Once these elementary things were sorted, I could run the test problem
example(mosek).

My question is: what happened to sol.itr.pobjval? It would be
convenient if this value was also
returned as part of the solution structure.

Thanks very much for all your efforts on Mosek and Rmosek,

RK

[Henrik Alsing Friberg]

Dear Roger,

Thank you for sharing your installation tips. In the new version 1.2.2
of Rmosek on CRAN, the option 'soldetail' has been added with default
value 0. If this is set to 1, 'pobjval' and 'dobjval' (or 'pobjbound'
in case of integer optimization) is returned as part of the solution
structure such that e.g. sol$itr$pobjval will be available. More
information can be retrieved by setting 'soldetail' to higher values,
or by setting the option 'getinfo' to TRUE. Please see the userguide
for more details.

Please note that this version also introduces the validation of
variable bounds and a simple check for disciplined convexity. Unsafe
variable bounds can cause operators to be evaluated outside their
domain and cause unexpected termination (that why it is at warning
level=3). Adding operators that are not disciplined convex will only
cause message to be printed at information level=4, because many
convex problems can be constructed without following these strict
rules. Use option 'verbose' to suppress this output stream.

Hope you find it useful!