Learning BN as probabilistic classifier

Learning a Bayesian network can be used to obtain a classifier for one of the nodes of the model. For more about classifier, see pyagrum.skbn.

Creative Commons License

aGrUM

interactive online version

 
In [1]:

import pyagrum as gum
import pyagrum.lib.notebook as gnb

from pyagrum.lib.bn2roc import showROC, showPR, showROC_PR

SIZE_LEARN = 10000
SIZE_VALID = 2000

In [2]:

bn = gum.loadBN("res/alarm.dsl")
bn

Out[2]:
G PAP PAP HREKG HREKG ANAPHYLAXIS ANAPHYLAXIS TPR TPR ANAPHYLAXIS->TPR DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE CATECHOL CATECHOL TPR->CATECHOL BP BP TPR->BP SAO2 SAO2 SAO2->CATECHOL HRSAT HRSAT CO CO CO->BP PCWP PCWP SHUNT SHUNT SHUNT->SAO2 INTUBATION INTUBATION INTUBATION->SHUNT VENTALV VENTALV INTUBATION->VENTALV VENTLUNG VENTLUNG INTUBATION->VENTLUNG MINVOL MINVOL INTUBATION->MINVOL PRESS PRESS INTUBATION->PRESS HR HR CATECHOL->HR VENTTUBE->VENTLUNG VENTTUBE->PRESS STROKEVOLUME STROKEVOLUME STROKEVOLUME->CO MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH HISTORY HISTORY PVSAT PVSAT VENTALV->PVSAT ARTCO2 ARTCO2 VENTALV->ARTCO2 INSUFFANESTH INSUFFANESTH INSUFFANESTH->CATECHOL FIO2 FIO2 FIO2->PVSAT CVP CVP VENTLUNG->VENTALV EXPCO2 EXPCO2 VENTLUNG->EXPCO2 VENTLUNG->MINVOL PULMEMBOLUS PULMEMBOLUS PULMEMBOLUS->PAP PULMEMBOLUS->SHUNT VENTMACH->VENTTUBE ERRLOWOUTPUT ERRLOWOUTPUT HRBP HRBP ERRLOWOUTPUT->HRBP PVSAT->SAO2 HR->HREKG HR->HRSAT HR->CO HR->HRBP HYPOVOLEMIA HYPOVOLEMIA HYPOVOLEMIA->STROKEVOLUME LVEDVOLUME LVEDVOLUME HYPOVOLEMIA->LVEDVOLUME ERRCAUTER ERRCAUTER ERRCAUTER->HREKG ERRCAUTER->HRSAT LVEDVOLUME->PCWP LVEDVOLUME->CVP ARTCO2->CATECHOL ARTCO2->EXPCO2 KINKEDTUBE KINKEDTUBE KINKEDTUBE->VENTLUNG KINKEDTUBE->PRESS LVFAILURE LVFAILURE LVFAILURE->STROKEVOLUME LVFAILURE->HISTORY LVFAILURE->LVEDVOLUME 
In [3]:

gum.generateSample(bn, SIZE_LEARN, "out/learn.csv", show_progress=True, with_labels=True)
gum.generateSample(bn, SIZE_VALID, "out/train.csv", show_progress=True, with_labels=True)

out/learn.csv: 100%|███████████████████████████████████████|

Log2-Likelihood : -150642.1695277766

out/train.csv: 100%|███████████████████████████████████████|

Log2-Likelihood : -30290.465260753102



Out[3]:

-30290.465260753102

Learning a BN from learn.csv

In [4]:

# Learning a BN from the database
learner = gum.BNLearner("out/train.csv")

bn2 = learner.useMIIC().learnBN()
currentTime = learner.currentTime()

In [5]:

gnb.flow.add(gnb.getBN(bn2, size="9"), f"Learned with {SIZE_LEARN} lines in {currentTime:.3f}s")
gnb.flow.display()
G PAP PAP PULMEMBOLUS PULMEMBOLUS PAP->PULMEMBOLUS HREKG HREKG SAO2 SAO2 CATECHOL CATECHOL SAO2->CATECHOL ANAPHYLAXIS ANAPHYLAXIS TPR TPR ANAPHYLAXIS->TPR BP BP TPR->BP HRSAT HRSAT HRSAT->HREKG DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE CO CO CO->BP PCWP PCWP SHUNT SHUNT SHUNT->SAO2 INTUBATION INTUBATION INTUBATION->SHUNT PRESS PRESS INTUBATION->PRESS HR HR CATECHOL->HR VENTLUNG VENTLUNG VENTTUBE->VENTLUNG VENTTUBE->PRESS STROKEVOLUME STROKEVOLUME STROKEVOLUME->CO MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH HISTORY HISTORY LVFAILURE LVFAILURE HISTORY->LVFAILURE VENTALV VENTALV VENTALV->INTUBATION MINVOL MINVOL VENTALV->MINVOL PVSAT PVSAT VENTALV->PVSAT ARTCO2 ARTCO2 VENTALV->ARTCO2 INSUFFANESTH INSUFFANESTH FIO2 FIO2 FIO2->PVSAT CVP CVP EXPCO2 EXPCO2 VENTLUNG->VENTALV VENTLUNG->EXPCO2 PULMEMBOLUS->SHUNT VENTMACH->VENTTUBE ERRLOWOUTPUT ERRLOWOUTPUT HRBP HRBP ERRLOWOUTPUT->HRBP HR->HREKG HR->HRSAT HR->CO HR->HRBP PVSAT->SAO2 HYPOVOLEMIA HYPOVOLEMIA HYPOVOLEMIA->STROKEVOLUME LVEDVOLUME LVEDVOLUME HYPOVOLEMIA->LVEDVOLUME LVEDVOLUME->PCWP LVEDVOLUME->CVP ERRCAUTER ERRCAUTER ERRCAUTER->HREKG ERRCAUTER->HRSAT ARTCO2->EXPCO2 KINKEDTUBE KINKEDTUBE KINKEDTUBE->VENTLUNG LVFAILURE->STROKEVOLUME LVFAILURE->LVEDVOLUME
Learned with 10000 lines in 0.259s
In [6]:

import pyagrum.lib.bn_vs_bn as bnvsbn

gnb.flow.add(gnb.getBNDiff(bn, bn2, size="8!"), "Diff with MIIC")
gnb.flow.add(bnvsbn.graphDiffLegend())
gnb.flow.display()
G KINKEDTUBE KINKEDTUBE VENTLUNG VENTLUNG KINKEDTUBE->VENTLUNG PRESS PRESS KINKEDTUBE->PRESS HYPOVOLEMIA HYPOVOLEMIA STROKEVOLUME STROKEVOLUME HYPOVOLEMIA->STROKEVOLUME LVEDVOLUME LVEDVOLUME HYPOVOLEMIA->LVEDVOLUME INTUBATION INTUBATION SHUNT SHUNT INTUBATION->SHUNT INTUBATION->VENTLUNG MINVOL MINVOL INTUBATION->MINVOL INTUBATION->PRESS VENTALV VENTALV MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH PULMEMBOLUS PULMEMBOLUS PAP PAP PULMEMBOLUS->SHUNT INSUFFANESTH INSUFFANESTH CATECHOL CATECHOL INSUFFANESTH->CATECHOL ERRLOWOUTPUT ERRLOWOUTPUT HRBP HRBP ERRLOWOUTPUT->HRBP ERRCAUTER ERRCAUTER HRSAT HRSAT ERRCAUTER->HRSAT HREKG HREKG ERRCAUTER->HREKG FIO2 FIO2 PVSAT PVSAT FIO2->PVSAT LVFAILURE LVFAILURE LVFAILURE->STROKEVOLUME LVFAILURE->LVEDVOLUME HISTORY HISTORY DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE ANAPHYLAXIS ANAPHYLAXIS TPR TPR ANAPHYLAXIS->TPR PAP->PULMEMBOLUS CO CO STROKEVOLUME->CO TPR->CATECHOL BP BP TPR->BP PCWP PCWP LVEDVOLUME->PCWP CVP CVP LVEDVOLUME->CVP VENTMACH->VENTTUBE SAO2 SAO2 SHUNT->SAO2 HISTORY->LVFAILURE VENTTUBE->VENTLUNG VENTTUBE->PRESS VENTLUNG->MINVOL VENTLUNG->VENTALV EXPCO2 EXPCO2 VENTLUNG->EXPCO2 VENTALV->INTUBATION VENTALV->MINVOL ARTCO2 ARTCO2 VENTALV->ARTCO2 VENTALV->PVSAT ARTCO2->EXPCO2 ARTCO2->CATECHOL PVSAT->SAO2 SAO2->CATECHOL HR HR CATECHOL->HR HR->HRBP HR->HRSAT HR->CO HR->HREKG HRSAT->HREKG CO->BP
Diff with MIIC
G a->b overflow c->d Missing e->f reversed g->h Correct 
In [7]:

bn3 = learner.useGreedyHillClimbing().useNMLCorrection().useScoreBDeu().learnBN()
gnb.flow.add(gnb.getBNDiff(bn, bn3, size="8!"), "Diff with GHC/NMD/BDEU")
gnb.flow.add(bnvsbn.graphDiffLegend())
gnb.flow.display()
G KINKEDTUBE KINKEDTUBE VENTLUNG VENTLUNG PRESS PRESS KINKEDTUBE->PRESS HYPOVOLEMIA HYPOVOLEMIA STROKEVOLUME STROKEVOLUME LVEDVOLUME LVEDVOLUME HYPOVOLEMIA->LVEDVOLUME PCWP PCWP HYPOVOLEMIA->PCWP INTUBATION INTUBATION SHUNT SHUNT INTUBATION->SHUNT MINVOL MINVOL INTUBATION->MINVOL INTUBATION->PRESS VENTALV VENTALV MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH PULMEMBOLUS PULMEMBOLUS PAP PAP PULMEMBOLUS->PAP PULMEMBOLUS->SHUNT INSUFFANESTH INSUFFANESTH INSUFFANESTH->PULMEMBOLUS CATECHOL CATECHOL INSUFFANESTH->CATECHOL ERRLOWOUTPUT ERRLOWOUTPUT HRBP HRBP ERRCAUTER ERRCAUTER HRSAT HRSAT HREKG HREKG FIO2 FIO2 PVSAT PVSAT FIO2->PVSAT LVFAILURE LVFAILURE HISTORY HISTORY LVFAILURE->HISTORY DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE ANAPHYLAXIS ANAPHYLAXIS TPR TPR PAP->ERRCAUTER STROKEVOLUME->HYPOVOLEMIA STROKEVOLUME->LVFAILURE STROKEVOLUME->HISTORY CO CO STROKEVOLUME->CO TPR->ANAPHYLAXIS BP BP TPR->BP LVEDVOLUME->LVFAILURE LVEDVOLUME->HISTORY CVP CVP LVEDVOLUME->CVP VENTMACH->VENTTUBE PCWP->LVEDVOLUME SHUNT->KINKEDTUBE SAO2 SAO2 SHUNT->SAO2 VENTTUBE->KINKEDTUBE VENTTUBE->INTUBATION VENTTUBE->VENTLUNG VENTTUBE->PRESS EXPCO2 EXPCO2 VENTTUBE->EXPCO2 VENTLUNG->KINKEDTUBE VENTLUNG->INTUBATION VENTLUNG->MINVOL VENTLUNG->VENTALV VENTALV->INTUBATION VENTALV->MINVOL ARTCO2 ARTCO2 VENTALV->ARTCO2 VENTALV->PVSAT ARTCO2->CATECHOL PVSAT->SAO2 SAO2->CATECHOL SAO2->HRBP EXPCO2->VENTLUNG EXPCO2->VENTALV EXPCO2->ARTCO2 CATECHOL->TPR HR HR HR->ERRLOWOUTPUT HR->ERRCAUTER HR->STROKEVOLUME HR->CATECHOL HR->HRSAT HR->CO HR->HREKG HRBP->ERRLOWOUTPUT HRBP->HR HRSAT->ERRCAUTER CO->BP HREKG->ERRCAUTER HREKG->HRSAT
Diff with GHC/NMD/BDEU
G a->b overflow c->d Missing e->f reversed g->h Correct 
In [8]:

bn4 = learner.useGreedyHillClimbing().useNMLCorrection().useScoreBDeu().setInitialDAG(bn2.dag()).learnBN()
gnb.flow.add(gnb.getBNDiff(bn, bn4, size="8!"), "Diff with GHC/NMD/BDEU with intial DAG from MIIC")
gnb.flow.add(bnvsbn.graphDiffLegend())
gnb.flow.display()
G KINKEDTUBE KINKEDTUBE VENTLUNG VENTLUNG KINKEDTUBE->VENTLUNG PRESS PRESS KINKEDTUBE->PRESS HYPOVOLEMIA HYPOVOLEMIA STROKEVOLUME STROKEVOLUME HYPOVOLEMIA->STROKEVOLUME LVEDVOLUME LVEDVOLUME HYPOVOLEMIA->LVEDVOLUME INTUBATION INTUBATION SHUNT SHUNT INTUBATION->SHUNT INTUBATION->VENTLUNG MINVOL MINVOL INTUBATION->MINVOL INTUBATION->PRESS VENTALV VENTALV INTUBATION->VENTALV MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH PULMEMBOLUS PULMEMBOLUS PAP PAP PULMEMBOLUS->PAP PULMEMBOLUS->SHUNT INSUFFANESTH INSUFFANESTH INSUFFANESTH->PULMEMBOLUS CATECHOL CATECHOL INSUFFANESTH->CATECHOL ERRLOWOUTPUT ERRLOWOUTPUT HRBP HRBP ERRLOWOUTPUT->HRBP ERRCAUTER ERRCAUTER HRSAT HRSAT ERRCAUTER->HRSAT HREKG HREKG ERRCAUTER->HREKG FIO2 FIO2 PVSAT PVSAT FIO2->PVSAT LVFAILURE LVFAILURE LVFAILURE->STROKEVOLUME LVFAILURE->LVEDVOLUME HISTORY HISTORY DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE ANAPHYLAXIS ANAPHYLAXIS TPR TPR CO CO STROKEVOLUME->CO TPR->ANAPHYLAXIS BP BP TPR->BP PCWP PCWP LVEDVOLUME->PCWP CVP CVP LVEDVOLUME->CVP VENTMACH->VENTTUBE SAO2 SAO2 SHUNT->SAO2 HISTORY->LVFAILURE VENTTUBE->INTUBATION VENTTUBE->VENTLUNG VENTTUBE->PRESS VENTLUNG->MINVOL VENTLUNG->VENTALV EXPCO2 EXPCO2 VENTLUNG->EXPCO2 VENTALV->MINVOL ARTCO2 ARTCO2 VENTALV->ARTCO2 VENTALV->PVSAT ARTCO2->EXPCO2 ARTCO2->CATECHOL PVSAT->SAO2 SAO2->CATECHOL CATECHOL->TPR HR HR CATECHOL->HR HR->HRBP HR->HRSAT HR->CO HR->HREKG HRSAT->HREKG CO->BP
Diff with GHC/NMD/BDEU with intial DAG from MIIC
G a->b overflow c->d Missing e->f reversed g->h Correct 
In [9]:

print(bn2.names())

{'PAP', 'HREKG', 'SAO2', 'ANAPHYLAXIS', 'TPR', 'HRSAT', 'DISCONNECT', 'CO', 'PCWP', 'SHUNT', 'INTUBATION', 'CATECHOL', 'BP', 'VENTTUBE', 'STROKEVOLUME', 'MINVOLSET', 'HISTORY', 'VENTALV', 'INSUFFANESTH', 'FIO2', 'CVP', 'EXPCO2', 'VENTLUNG', 'PULMEMBOLUS', 'MINVOL', 'VENTMACH', 'ERRLOWOUTPUT', 'HR', 'PVSAT', 'HYPOVOLEMIA', 'LVEDVOLUME', 'ERRCAUTER', 'ARTCO2', 'KINKEDTUBE', 'LVFAILURE', 'HRBP', 'PRESS'}

In [10]:

gnb.showInference(bn2, evs={}, size="10")
../_images/notebooks_32-Learning_learningClassifier_13_0.svg

Two classifiers from the learned BN

In [11]:

print(bn2["HRSAT"])
print(bn2["INTUBATION"])

HRSAT:Labelized({HIGH|LOW|NORMAL})
INTUBATION:Labelized({ESOPHAGEAL|NORMAL|ONESIDED})

In [12]:

showROC(bn2, "out/train.csv", "HRSAT", "LOW", show_progress=False)
showROC(bn2, "out/train.csv", "HRSAT", "NORMAL", show_progress=False)
showROC(bn2, "out/train.csv", "HRSAT", "HIGH", show_progress=False);
../_images/notebooks_32-Learning_learningClassifier_16_0.svg
../_images/notebooks_32-Learning_learningClassifier_16_1.svg
../_images/notebooks_32-Learning_learningClassifier_16_2.svg 
In [13]:

showROC(bn2, "out/train.csv", "INTUBATION", "ESOPHAGEAL", show_progress=False);
../_images/notebooks_32-Learning_learningClassifier_17_0.svg 
In [14]:

showPR(bn2, "out/train.csv", "HRSAT", "LOW", show_progress=False);
../_images/notebooks_32-Learning_learningClassifier_18_0.svg 
In [15]:

showPR(bn2, "out/train.csv", "INTUBATION", "ESOPHAGEAL", show_progress=False);
../_images/notebooks_32-Learning_learningClassifier_19_0.svg 
In [16]:

showROC_PR(bn2, "out/train.csv", "HRSAT", "LOW", show_progress=False);
../_images/notebooks_32-Learning_learningClassifier_20_0.svg 
In [17]:

showROC_PR(bn2, "out/train.csv", "INTUBATION", "ESOPHAGEAL", show_progress=False);
../_images/notebooks_32-Learning_learningClassifier_21_0.svg