Bug in free format read

[jajhall]

I've found two bugs in the free format read routine. They show up because HiGHS using the free format MPS read doesn't get the optimal objective value obtained by Hsol, Clp, Glop and Soplex - so it's an MPS read problem, not a solver issue!

Bug 1

For mod2 (at least) there are multiple "N" rows in the MPS file

It looks as if the FF read merges them all. Specifically

The first "N" Row is called "func", there is another called "noipr.ad"

Column "aleo....aa" has no entry in the row "func" - so Hsol etc give it a cost coefficient of zero. However, it has an entry of 1.0 in row "noipr.ad", and the free format read gives a cost coefficient of 1.0

The constraint matrix read is also compromised - possibly for the same reason. The FF read yields 198250 entries, whereas the matrix from Hsol etc has 165129 entries.

Bug 2

For dcp2

http://www.maths.ed.ac.uk/hall/PublicLP/

the FF read fails to get the row upper bounds correct: making them infinite rather than the finite value which comes as a result of the RANGES section.

I hope that these can be fixed soon, since I need to be confident that models are being read correctly when benchmarking and analysing models via the behaviour of HiGHS.

[galabovaa]

dcp2 issue fixed.

For mod2 I now get the opposite:

Free format yields 165129 because it doesn't add the free rows to the constraint matrix but fixed format read yields 198250 because it adds them. To see if this causes the issue I modified free format read to add them as well. Sadly, this did not change anything - we still get a different objective value. I am not sure why yet, the bug must be elsewhere.

[jajhall]

Strange. I can't investigate this on the bus... Julian On 7 March 2019 16:32:52 GMT+00:00, galabovaa <notifications@github.com> wrote: dcp2 issue fixed. For mod2 I now get the opposite: Free format yields 165129 because it doesn't add the free rows to the constraint matrix but fixed format read yields 198250 because it adds them. To see if this causes the issue I modified free format read to add them as well. Sadly, this did not change anything - we still get a different objective value. I am not sure why yet, the bug must be elsewhere. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<#20 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AVll4wsPKU9Ea4PQXsp_XSjen7o3JeNBks5vUT80gaJpZM4V_lcp>. Julian Hall -- Dr. J. A. Julian Hall Reader, School of Mathematics University of Edinburgh Email: J.A.J.Hall@ed.ac.uk Web: http://www.maths.ed.ac.uk/people/show/person/47 The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

[galabovaa]

mod2 now solves the same with both parsers.