Problem with column names containing space characters

[mocenigo]

I have some code that uses

SB = model.addBinaries(rounds, slices, columns, rows, name_prefix = 'SB')

where rounds, slices, columns, rows are lists, and this generates variables with names such as SB(0, 0, 0, 0).

I then proceed to write the model to a file

model.writeModel(tmp_model_file_path)

but at this point I get the error

ERROR:   Column 0 name "SB(0, 0, 0, 0)" contains a space character

which, I guess, should not happen, since it has been generated by HiGHS itself...

[mocenigo]

I have made this work by removing the spaces instead.
I am not sure I have written a really bug free code, but it works for now
from around line 416 of file highs/lp_data/HighsModelUtils.cpp

    } else if (names[ix].find(" ") != std::string::npos) {
        // Remove spaces from the name
        names[ix].erase(std::remove(names[ix].begin(), names[ix].end(), ' '), names[ix].end());
    }

[jajhall]

Thanks, we need to ensure that addBinaries does not insert spaces into names

[mocenigo]

Thanks, we need to ensure that addBinaries does not insert spaces into names

I could have been megalomaniac and could have made a tiny PR but I guess this is faster :-)
I am not a c++ programmer, in fact I loathe the language, so I do not know how idiomatic my one-liner is, but I guess it is ok.

[jajhall]

Rather than remove the spaces within HiGHS - which leads to names differing from what's defined by the user - it's better to prevent them in the first place.

I thought it would be more obvious where/how the spaces are created by highs.py, but I can't see.

I am not a Python programmer, in fact I loathe the language...

@mathgeekcoder, can you help here?

[mocenigo]

I am not a Python programmer, in fact I loathe the language...

Hehe, chacun à son goût...

[mocenigo]

Rather than remove the spaces within HiGHS - which leads to names differing from what's defined by the user - it's better to prevent them in the first place.

I thought it would be more obvious where/how the spaces are created by highs.py, but I can't see.

I am not a Python programmer, in fact I loathe the language...

@mathgeekcoder, can you help here?

Well, in 1.11 and earlier, addVariables was still generating the names with the spaces, but the code did not mark it as an error. It just parsed things like SB(0, 3, 2, 4) as a single name.

[jajhall]

Yes, in 1.11 and earlier, HiGHS would write out LP and MPS files with spaces in variable and constraint names that it would then be unable to read in. Hence we switched to returning an error if the name contained spaces.

[mocenigo]

The issue seems to be with the implementation of product

file highspy.py, line 648

        indices: list[Any] = list(expanded[0] if len(expanded) == 1 else product(*expanded))  # unpack tuple if needed

and later, line 743

            names = name or [f"{name_prefix}{i}" for i in indices]

and it is THIS conversion of the indices (or multi-indices) to strings that preserves spaces, because this is the way they are printed...

Hence, if you change line 743 to

            names = name or [f"{name_prefix}{i}".replace(" ", "")  for i in indices]

then this is fixed (it also removes any spaces from "name_prefix").

[mathgeekcoder]

Yes, @mocenigo has the right idea for the fix! I'm happy to submit a PR, unless @mocenigo wants to do the honours.

[jajhall]

The issue seems to be with the implementation of product

file highspy.py, line 648

        indices: list[Any] = list(expanded[0] if len(expanded) == 1 else product(*expanded))  # unpack tuple if needed

and later, line 743

            names = name or [f"{name_prefix}{i}" for i in indices]

and it is THIS conversion of the indices (or multi-indices) to strings that preserves spaces, because this is the way they are printed...

Hence, if you change line 743 to

            names = name or [f"{name_prefix}{i}".replace(" ", "")  for i in indices]

then this is fixed (it also removes any spaces from "name_prefix").

I feel that the spaces should be replaced by underscores, since the spaces were inserted to separate the terms

[mocenigo]

The issue seems to be with the implementation of product
file highspy.py, line 648

        indices: list[Any] = list(expanded[0] if len(expanded) == 1 else product(*expanded))  # unpack tuple if needed

and later, line 743

            names = name or [f"{name_prefix}{i}" for i in indices]

and it is THIS conversion of the indices (or multi-indices) to strings that preserves spaces, because this is the way they are printed...
Hence, if you change line 743 to

            names = name or [f"{name_prefix}{i}".replace(" ", "")  for i in indices]

then this is fixed (it also removes any spaces from "name_prefix").

I feel that the spaces should be replaced by underscores, since the spaces were inserted to separate the terms

No, because there are already commas.
So, instead of the strings

'MC(0, 0, 0, 0)', 'MC(0, 0, 0, 1)', 'MC(0, 0, 0, 2)', ...

for the column names, you now get

'MC(0,0,0,0)', 'MC(0,0,0,1)', 'MC(0,0,0,2)', ...