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.bgum")
bn

Out[2]:
G INTUBATION INTUBATION PRESS PRESS INTUBATION->PRESS MINVOL MINVOL INTUBATION->MINVOL SHUNT SHUNT INTUBATION->SHUNT VENTLUNG VENTLUNG INTUBATION->VENTLUNG VENTALV VENTALV INTUBATION->VENTALV FIO2 FIO2 PVSAT PVSAT FIO2->PVSAT PULMEMBOLUS PULMEMBOLUS PULMEMBOLUS->SHUNT PAP PAP PULMEMBOLUS->PAP LVFAILURE LVFAILURE STROKEVOLUME STROKEVOLUME LVFAILURE->STROKEVOLUME LVEDVOLUME LVEDVOLUME LVFAILURE->LVEDVOLUME HISTORY HISTORY LVFAILURE->HISTORY PCWP PCWP HR HR CO CO HR->CO HRSAT HRSAT HR->HRSAT HRBP HRBP HR->HRBP HREKG HREKG HR->HREKG ERRLOWOUTPUT ERRLOWOUTPUT ERRLOWOUTPUT->HRBP HYPOVOLEMIA HYPOVOLEMIA HYPOVOLEMIA->STROKEVOLUME HYPOVOLEMIA->LVEDVOLUME CATECHOL CATECHOL CATECHOL->HR SAO2 SAO2 PVSAT->SAO2 MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH BP BP CO->BP ARTCO2 ARTCO2 ARTCO2->CATECHOL EXPCO2 EXPCO2 ARTCO2->EXPCO2 DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE SHUNT->SAO2 KINKEDTUBE KINKEDTUBE KINKEDTUBE->PRESS KINKEDTUBE->VENTLUNG VENTLUNG->MINVOL VENTLUNG->EXPCO2 VENTLUNG->VENTALV STROKEVOLUME->CO LVEDVOLUME->PCWP CVP CVP LVEDVOLUME->CVP INSUFFANESTH INSUFFANESTH INSUFFANESTH->CATECHOL TPR TPR TPR->CATECHOL TPR->BP ANAPHYLAXIS ANAPHYLAXIS ANAPHYLAXIS->TPR VENTMACH->VENTTUBE VENTTUBE->PRESS VENTTUBE->VENTLUNG SAO2->CATECHOL VENTALV->PVSAT VENTALV->ARTCO2 ERRCAUTER ERRCAUTER ERRCAUTER->HRSAT ERRCAUTER->HREKG 
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 : -151669.28259683488

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

Log2-Likelihood : -30238.359952531624

Out[3]:

-30238.359952531624

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 INTUBATION INTUBATION PRESS PRESS INTUBATION->PRESS SHUNT SHUNT INTUBATION->SHUNT VENTALV VENTALV INTUBATION->VENTALV FIO2 FIO2 PVSAT PVSAT FIO2->PVSAT PULMEMBOLUS PULMEMBOLUS PAP PAP PULMEMBOLUS->PAP LVFAILURE LVFAILURE STROKEVOLUME STROKEVOLUME LVFAILURE->STROKEVOLUME LVEDVOLUME LVEDVOLUME LVFAILURE->LVEDVOLUME PCWP PCWP HR HR CO CO HR->CO HRSAT HRSAT HR->HRSAT HRBP HRBP HR->HRBP HREKG HREKG HR->HREKG ERRLOWOUTPUT ERRLOWOUTPUT ERRLOWOUTPUT->HRBP HYPOVOLEMIA HYPOVOLEMIA HYPOVOLEMIA->STROKEVOLUME CATECHOL CATECHOL CATECHOL->HR SAO2 SAO2 PVSAT->SAO2 MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH BP BP CO->BP ARTCO2 ARTCO2 EXPCO2 EXPCO2 ARTCO2->EXPCO2 MINVOL MINVOL MINVOL->INTUBATION DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE SHUNT->PULMEMBOLUS SHUNT->SAO2 KINKEDTUBE KINKEDTUBE VENTLUNG VENTLUNG KINKEDTUBE->VENTLUNG VENTLUNG->MINVOL VENTLUNG->EXPCO2 VENTLUNG->VENTALV HRSAT->HREKG STROKEVOLUME->CO LVEDVOLUME->PCWP LVEDVOLUME->HYPOVOLEMIA CVP CVP LVEDVOLUME->CVP HISTORY HISTORY HISTORY->LVFAILURE INSUFFANESTH INSUFFANESTH TPR TPR TPR->BP ANAPHYLAXIS ANAPHYLAXIS ANAPHYLAXIS->TPR VENTMACH->VENTTUBE SAO2->CATECHOL VENTTUBE->PRESS VENTTUBE->VENTLUNG VENTALV->PVSAT VENTALV->ARTCO2 ERRCAUTER ERRCAUTER ERRCAUTER->HRSAT ERRCAUTER->HREKG
Learned with 10000 lines in 1.965s
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 INTUBATION INTUBATION SHUNT SHUNT INTUBATION->SHUNT INTUBATION->VENTLUNG MINVOL MINVOL INTUBATION->PRESS VENTALV VENTALV INTUBATION->VENTALV MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH PULMEMBOLUS PULMEMBOLUS PAP PAP PULMEMBOLUS->PAP 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 CO CO STROKEVOLUME->CO TPR->CATECHOL BP BP TPR->BP LVEDVOLUME->HYPOVOLEMIA PCWP PCWP LVEDVOLUME->PCWP CVP CVP LVEDVOLUME->CVP VENTMACH->VENTTUBE SHUNT->PULMEMBOLUS SAO2 SAO2 SHUNT->SAO2 HISTORY->LVFAILURE VENTTUBE->VENTLUNG VENTTUBE->PRESS VENTLUNG->MINVOL VENTLUNG->VENTALV EXPCO2 EXPCO2 VENTLUNG->EXPCO2 MINVOL->INTUBATION 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 HYPOVOLEMIA->STROKEVOLUME LVEDVOLUME LVEDVOLUME INTUBATION INTUBATION SHUNT SHUNT INTUBATION->SHUNT MINVOL MINVOL INTUBATION->MINVOL INTUBATION->PRESS VENTALV VENTALV MINVOLSET MINVOLSET MINVOLSET->INTUBATION VENTMACH VENTMACH MINVOLSET->VENTMACH PULMEMBOLUS PULMEMBOLUS PAP PAP PULMEMBOLUS->PAP PVSAT PVSAT PULMEMBOLUS->PVSAT INSUFFANESTH INSUFFANESTH CATECHOL CATECHOL INSUFFANESTH->CATECHOL ERRLOWOUTPUT ERRLOWOUTPUT HR HR ERRLOWOUTPUT->HR HRBP HRBP ERRCAUTER ERRCAUTER HRSAT HRSAT ERRCAUTER->HRSAT HREKG HREKG ERRCAUTER->HREKG FIO2 FIO2 FIO2->LVEDVOLUME FIO2->PVSAT LVFAILURE LVFAILURE LVFAILURE->STROKEVOLUME HISTORY HISTORY LVFAILURE->HISTORY DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE ANAPHYLAXIS ANAPHYLAXIS TPR TPR CO CO STROKEVOLUME->CO TPR->ANAPHYLAXIS BP BP TPR->BP LVEDVOLUME->HYPOVOLEMIA LVEDVOLUME->LVFAILURE PCWP PCWP CVP CVP LVEDVOLUME->CVP VENTMACH->VENTTUBE PCWP->LVEDVOLUME SHUNT->KINKEDTUBE SHUNT->PULMEMBOLUS SAO2 SAO2 SHUNT->SAO2 VENTTUBE->KINKEDTUBE VENTTUBE->VENTLUNG VENTTUBE->PRESS VENTLUNG->KINKEDTUBE VENTLUNG->INTUBATION VENTLUNG->MINVOL VENTLUNG->VENTALV EXPCO2 EXPCO2 VENTLUNG->EXPCO2 VENTALV->INTUBATION ARTCO2 ARTCO2 VENTALV->ARTCO2 VENTALV->PVSAT ARTCO2->EXPCO2 ARTCO2->CATECHOL PVSAT->SAO2 SAO2->CATECHOL SAO2->HRBP CATECHOL->TPR HR->CATECHOL HR->HRSAT HR->CO HR->HREKG HRBP->ERRLOWOUTPUT HRBP->HR CO->BP
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 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 PVSAT PVSAT PULMEMBOLUS->PVSAT INSUFFANESTH INSUFFANESTH CATECHOL CATECHOL INSUFFANESTH->CATECHOL ERRLOWOUTPUT ERRLOWOUTPUT HRBP HRBP ERRLOWOUTPUT->HRBP ERRCAUTER ERRCAUTER HRSAT HRSAT ERRCAUTER->HRSAT HREKG HREKG ERRCAUTER->HREKG FIO2 FIO2 PCWP PCWP FIO2->PCWP FIO2->PVSAT LVFAILURE LVFAILURE LVFAILURE->STROKEVOLUME LVFAILURE->LVEDVOLUME HISTORY HISTORY DISCONNECT DISCONNECT VENTTUBE VENTTUBE DISCONNECT->VENTTUBE ANAPHYLAXIS ANAPHYLAXIS TPR TPR ANAPHYLAXIS->TPR CO CO STROKEVOLUME->CO TPR->CATECHOL BP BP TPR->BP LVEDVOLUME->HYPOVOLEMIA LVEDVOLUME->FIO2 LVEDVOLUME->PCWP CVP CVP LVEDVOLUME->CVP VENTMACH->VENTTUBE SHUNT->PULMEMBOLUS SAO2 SAO2 SHUNT->SAO2 HISTORY->INTUBATION HISTORY->LVFAILURE VENTTUBE->VENTLUNG VENTTUBE->PRESS VENTLUNG->MINVOL VENTLUNG->VENTALV EXPCO2 EXPCO2 VENTLUNG->EXPCO2 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 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())

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

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