Is it possible to read a solution file created using highspy in one format or style, and convert it to another format without the need to run the model?

[hbshrestha]

Hello. I ran a model and created a solution file using style 1 with the following code a week ago.

import highspy

h = highspy.Highs()
filename = 'abc.mps'
h.readModel(filename)

h.setOptionValue("solver","ipm")
h.run()

h.writeSolution("abc_highs_solution1.sol", 1)

I learnt yesterday #2072 that it is also possible to write the solution file in different format using HIGHS. As per the documentation, style 1 refers to HIGHS pretty format, and style 3 refers to Glpsol pretty format.

I need the solution file in style 3. The model had taken two days to run. Therefore, I'd like to avoid running the model again if at all possible. 1. Is it possible to read the abc_highs_solution1.sol file using highspy and convert it into style 3 directly without requiring to run the model again?

I have a second sub-question.
2. In the style 3 solution file, can the index number go beyond six digits (to 1 million and further)?

[jajhall]

This should work.

If the problem is an LP, call run() and HiGHS should just validate that the solution read in is optimal, and then write out in Glpsol format.

If it's a MIP, it's less clear. Don't call run(), but hope that it writes out the solution, despite not having the model status set.

It would have been sensible to check that this mechanism works before solving such a large problem, but I know life's not like that at times!

[hbshrestha]

Thank you for the response! It is an LP problem.

I noticed it is possible to read solution. When I tried,

h = highspy.Highs()
filename = "abc_highs_solution1.sol"
h.readsolution("abc_highs_solution1.sol", 1)

I got <HighsStatus.kError: -1>.

One has to pass arg0 and arg1. But, what do they refer to?

[hbshrestha]

Thank you. Please take your time.

[jajhall]

Unfortunately - and I've looked back properly at your earlier questions

As per the documentation, style 1 refers to HIGHS old raw format, and style 3 refers to Glpsol pretty format.

write_solution_style = 1 is not a HIGHS raw format (-1 => old raw; 0 => current raw), but HiGHS "pretty" format. HiGHS can't read its pretty format, as this is for humans.

So, you'll have to convert the HiGHS "pretty" format file into Glpsol "pretty" format using some text processing package (like perl/awk or some modern equivalent)

2. In the style 3 solution file, can the index number go beyond six digits (to 1 million and further)?

Yes, up to 9 digits!

In passing, there is at least one bug in the code to write out in Glpsol "pretty" format. I get lines like

No. Row name St Activity Lower bound Upper bound Marginal

---

 1 .Z....       B         225495                             
 2 ....01       NU             0                           0         -3310 
 3 ....02       NS          52.6          52.6             =       775.224 

I can look into this and probably soon fix it

[jajhall]

Yes, I saw that, too. I think I know the cause, and it should be simple to fix.

[jajhall]

Besides, the = sign, I also noticed "<eps" mentioned in Marginal column. I am not sure what it means and if it is a bug.

Looking at the code, writing "=" for the upper bound when it's equal to the lower bound, and "< eps" for the "Marginal" when it is less than 1e-9, is clearly hard-coded, so I have to assume that it's the correct format for the "pretty" Glpsol file. When I wrote the Glpsol output utility, I was studying the Glpsol source code to see how it handled these cases.

[hbshrestha]

As per your response, I'd understand there is not a bug. "=" is just a different way of denoting same value as bounds, and "<eps" would represent less than 1e-9. So, we can still use the solution file, directly, right?

[jajhall]

Yes, I correctly reproduced the glpsol style in the pretty format

[hbshrestha]

Thank you.

[jajhall]

By passing 3, I am getting very similar to the GLPK format I shared in the question. Header names are different.
h.writeSolution("abc_highs_solution.sol", 3)

The format I created was for Glpsol, not GLPK. I know that they are closely related, and if the header files are different, I can try to add GLPK raw and pretty format - as I'm keen to wean people off Glpsol/GLPK since their solvers have such poor performance.

Can Glpsol/GLPK read their own pretty format?

[hbshrestha]

I could still use the style 3 solution file for getting results in my small model tests.
I am not sure whether Glpsol/GLPK can read their own pretty format.