MUS extraction unsupported for CNF with either OptUX and MUSx ? about pysat

[alexeyignatiev]

Hi @fransab,

The documentation should say (see the list of expected arguments and their types) that the input parameters to MUSX and OptUx should be weighted formulas. Namely, MUSX expects to get a WCNF object while OptUx wants WCNFPlus object. To make your example work, you can make a tiny modification (a call to .weighted(), which will create a weighted variant of the formula for you):

>>> from pysat.examples.musx import MUSX
>>> cnf = CNF(from_clauses=[[1, 2], [-1], [-2]])
>>> musx = MUSX(cnf.weighted(), verbosity=0)
>>> musx.compute()
[1, 2, 3]
>>> musx.delete()
>>> with OptUx(cnf.weighted()) as optux:
...     for mus in optux.enumerate():
...         print('mus {0} has cost {1}'.format(mus, optux.cost))
...
mus [1, 2, 3] has cost 3

Note that you can call either of these tools on an unweighted *.cnf file because the script will call .weighted() for you after parsing the file.

As for the WCNF file you are creating, it is incorrect. If you want each clause to be considered as a culprit for unsatisfiability then each clause should be marked as soft. This means you should set the top weight (in the preamble) to the sum of weights for all soft clauses + 1, i.e. it should be 4 rather than 1. Currently, all the clauses in the file are made hard, hence the cores MUSX and OptUx are getting are empty.

I hope this helps,
Alexey

[fransab]

Thanks for your help @alexeyignatiev .
I guess it would be nice for these MUS extractors to directly support formulas in CNF formulas, given the minimal modification required, unless specific situations could return an unwanted result ?
Sorry for my misunderstanding of the WCNF format and thanks for your clarification, I understand why it works now.

Francesco