Precision of floats in lp files about linopy

[FabianHofmann]

Hey @jafluri, thanks for pointing this out. Correct me, if I am wrong, but the default precision seems to be 6. Since we do not want to bloat the LP file size, could you try out whether a precision of 8/9 is already satisfying?

[jafluri]

Thanks for the quick reply. I just ran the model for 6, 9, 12 and 15 significant digits. The objective for 9, 12 and 15 is identical. Comparing the variable values to the ones from pyomo, you get the smallest differences with 12 and 15, but I guess 9 should be enough as some floating point jitter is to be expected in large models.

[jafluri]

In combination with #112, one could do "%+.12g", that strips away the tailing zeros:

array([1. , 1.2 , 1.2334567]) -> array(['+1', '+1.2', '+1.2334567'], dtype=object)

I'd assume that way you'd safe memory in most cases, even if you go up to 12, compared to the current implementation.

[FabianHofmann]

Oh, this sounds great! Could you do a PR?

[jafluri]

Thanks for the quick resolution, I'm closing this then.