Presolve error

[mckib2]

Related to #13 , possibly #280

Solving simple problem:

c = [-3. -2.]
A=   (0, 0)     2.0
  (1, 0)        1.0
  (2, 0)        1.0
  (0, 1)        1.0
  (1, 1)        1.0
rhs= [10.  8.  4.]
lhs= [-inf -inf -inf]
lb= [0. 0.]
ub= [inf inf]
options= {'presolve': True, 'sense': 1, 'solver': 'simplex', 'parallel': False, 'time_limit': 1, 'message_level': 1}

results in the error message:

terminate called after throwing an instance of 'std::out_of_range'
  what():  vector::_M_range_check: __n (which is 2) >= this->size() (which is 2)
Aborted (core dumped)

Solution is incorrectly reported when presolve=False as [0, 8], should be [2, 6].

However, [2, 6, (undefined)] is the reported answer when transpose(A) is used, with and without presolve enabled (sometimes fails because 3rd value is undefined). Is there something I'm not understanding about the interface? Because the dimensions of A don't seem like they work like that.

[galabovaa]

It is not that you are missing something, we have bugs we need to clear out and your examples are very helpful. Lots of changes have happened recently and we need to do some bug fixing, I will look into the example asap. Again, better docs will be available soon. Until then yes, the infinity value may cause a problem.

[galabovaa]

Why do you think it is linked to #13? The problems causing the issue there presolve OK and I just ran the code again and got no errors.

[mckib2]

Error message from #13 is the same, could be that it is not related. Do you have an idea of which infinity values may be causing the issue?

[mckib2]

Please let me know if there's a better way for me to report problems, is MPS better or just a description like above sufficient?

[galabovaa]

Thank you, our tests (including interface tests) are WIP so if you have an mps available that would be great!

@jajhall Do you know which infinity values may cause problems? I remember there was a case like that but not the details and I don't know whether you fixed that already.

[jajhall]

Related to #13 , possibly #280

Solving simple problem:

c = [-3. -2.]
A=   (0, 0)     2.0
  (1, 0)        1.0
  (2, 0)        1.0
  (0, 1)        1.0
  (1, 1)        1.0
rhs= [10.  8.  4.]
lhs= [-inf -inf -inf]
lb= [0. 0.]
ub= [inf inf]
options= {'presolve': True, 'sense': 1, 'solver': 'simplex', 'parallel': False, 'time_limit': 1, 'message_level': 1}

results in the error message:

terminate called after throwing an instance of 'std::out_of_range'
  what():  vector::_M_range_check: __n (which is 2) >= this->size() (which is 2)
Aborted (core dumped)

Solution is incorrectly reported when presolve=False as [0, 8], should be [2, 6].

However, [2, 6, (undefined)] is the reported answer when transpose(A) is used, with and without presolve enabled (sometimes fails because 3rd value is undefined). Is there something I'm not understanding about the interface? Because the dimensions of A don't seem like they work like that.

You can't just transpose A since its dimensions must be consistent with c, rhs, lhs, lb and ub

That said, I don't see how you get from the problem definition you give, and HiGHS. What is the value of inf? Or, more precisely, what value for inf reaches HiGHS?

[mckib2]

You can't just transpose A since its dimensions must be consistent with c, rhs, lhs, lb and ub

This is why I am confused about results.

That said, I don't see how you get from the problem definition you give, and HiGHS. What is the value of inf? Or, more precisely, what value for inf reaches HiGHS?

Value of inf is HIGHS_CONST_INF. The problem is passed through the Cython wrapper scikit-highs that we are working on for inclusion in scipy. scikit-highs is built on top of the C API.

I will generate an MPS file that describes the problem later today.

[jajhall]

You've got a fault because you're not using the up-to-date version of the C API. When I was improving HiGHS/examples/call_highs_from_c.c the other day, I decided to use an example with 3 rows and 2 columns, and got a fault due to the following lines in src/interfaces/highs_c_api.cpp

lp.rowLower_.assign(rowlower, rowlower + numrow);
lp.rowUpper_.assign(rowupper, rowupper + numcol);

The second should say "+ numrow);"

[mckib2]

I just pulled in all the recent commits from master, C API now provides the expected results. Thank you both for helping me troubleshoot that!