Table 1.
Prior algorithms for learning multiple Markov boundaries and variable sets. “+” means that the corresponding property is satisfied by a method, “−” means that the property is not satisfied, and “+/−” denotes cases where the property is satisfied under certain conditions.
| Markov boundary identification (assuming faithfulness except for violations of the intersection property) | Parameterization: does not require prior knowledge of | Computationally efficient | sample efficient | |||
|---|---|---|---|---|---|---|
| correct (identifies Markov boundaries) | complete (identifies all Markov boundaries) | the number of Markov boundaries/variable sets | the size of Markov boundaries/variable sets | |||
| KIAMB | + | + | − | + | − | − |
| EGS-CMIM | − | − | − | − | − | + |
| EGS-NCMIGS | − | − | − | +/− | − | + |
| EGSG | − | − | − | + | − | + |
| Resampling+RFE | − | − | − | + | − | + |
| Resampling+UAF | − | − | − | + | − | + |
| IR-HITON-PC | + | − | + | + | + | + |
| IR-SPLR | − | − | + | + | + | + |