Statistics
Tests over valid formulas and over random formulas
The performances of PRONOM seems prmising. We have tested it:
- on sets of valid formulas for each modal logic
- on sets of randomly generated formulas, obtained by fixing the number of propositional variables,
the depth (level of nesting of connectives) of a formula and a time limit.
To run tests:
- consult the Prolog source file of the system to be tested (e.g. e.pl)
- consult the Prolog source file of statistics
- to run tests over valid formulas, ask the Prolog engine to prove
runValidFormulas(+system,+time_limit_in_seconds)
where system can be: e, em, en, ec, enc, emn, emc, emnc
- to run tests over randomly generated formulas, ask the Prolog engine to prove
runRandomFormulas(+number_of_tests,+number_of_propositional_variables,+depth,+time_limit_in_seconds)
Results of tests over valid formulas are shown in Table 1, whereas the percentage
of timeouts in executing tests over randomly generated formulas are shown in Table 2.

Table 1: percentage of timeouts of PRONOM over N valid formulas, 90 < N < 500.

Table 2: percentage of timeouts of PRONOM over 8000 formulas randomly generated.