Biology Reference
In-Depth Information
(a) The model string reported in the manual page is the following:
> true = empty.graph(names(alarm))
> modelstring(true) = paste("[HIST|LVF][CVP|LVV]",
+ "[PCWP|LVV][HYP][LVV|HYP:LVF][LVF]",
+ "[STKV|HYP:LVF][ERLO][HRBP|ERLO:HR]",
+ "[HREK|ERCA:HR][ERCA][HRSA|ERCA:HR][ANES]",
+ "[APL][TPR|APL][ECO2|ACO2:VLNG][KINK]",
+ "[MINV|INT:VLNG][FIO2][PVS|FIO2:VALV]",
+ "[SAO2|PVS:SHNT][PAP|PMB][PMB][SHNT|INT:PMB]",
+ "[INT][PRSS|INT:KINK:VTUB][DISC][MVS]",
+ "[VMCH|MVS][VTUB|DISC:VMCH]",
+ "[VLNG|INT:KINK:VTUB][VALV|INT:VLNG]",
+ "[ACO2|VALV][CCHL|ACO2:ANES:SAO2:TPR]",
+ "[HR|CCHL][CO|HR:STKV][BP|CO:TPR]", sep = "")
(b) The performance of constraint learning algorithms for the alarm data set im-
proves in newer algorithms, which is not unexpected because this network is
used so often as a benchmark in publications. However, note how with the de-
fault parameters the network structures that are learned from the data differ from
true . The overall number of arcs in each network is similar to the number of
arcs in true , but many false positives and false negatives are present. As a re-
sult, network scores for the learned networks are much lower than the one for
thetruenetwork.
> bn.gs = gs(alarm)
> bn.iamb = iamb(alarm)
> bn.inter = inter.iamb(alarm)
> par(mfrow = c(2, 2))
> graphviz.plot(true, main = "True Structure")
> graphviz.plot(bn.gs, main = "Grow-Shrink")
> graphviz.plot(bn.iamb, main = "IAMB")
> graphviz.plot(bn.inter, main = "Inter-IAMB")
> unlist(compare(true, bn.gs))
> unlist(compare(true, bn.iamb))
> unlist(compare(true, bn.inter))
> score(cextend(bn.gs), alarm, type = "bde")
> score(cextend(bn.iamb), alarm, type = "bde")
> score(cextend(bn.inter), alarm, type = "bde")
(c) Permutation tests significantly improve the performance of both IAMB and
Inter-IAMB: true positives (i.e., arcs correctly included in the network) increase
for all learning algorithms. False positives (i.e., arcs incorrectly included in
the network) and false negatives (i.e., arcs incorrectly absent from the network)
are also fewer, though by a smaller amount.
Search WWH ::




Custom Search