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 sys
import os

import numpy as np

import pyAgrum as gum
import pyAgrum.lib.notebook as gnb

%matplotlib inline
from pyAgrum.lib.bn2roc import showROC
from pyAgrum.lib.bn2roc import showPR
from pyAgrum.lib.bn2roc import showROC_PR

SIZE_LEARN=10000
SIZE_VALID=2000
In [2]:
bn=gum.loadBN("res/alarm.dsl")
bn
Out[2]:
G PULMEMBOLUS PULMEMBOLUS PAP PAP PULMEMBOLUS->PAP SHUNT SHUNT PULMEMBOLUS->SHUNT ANAPHYLAXIS ANAPHYLAXIS TPR TPR ANAPHYLAXIS->TPR MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH HYPOVOLEMIA HYPOVOLEMIA LVEDVOLUME LVEDVOLUME HYPOVOLEMIA->LVEDVOLUME STROKEVOLUME STROKEVOLUME HYPOVOLEMIA->STROKEVOLUME HRSAT HRSAT MINVOL MINVOL INTUBATION INTUBATION INTUBATION->MINVOL VENTLUNG VENTLUNG INTUBATION->VENTLUNG PRESS PRESS INTUBATION->PRESS VENTALV VENTALV INTUBATION->VENTALV INTUBATION->SHUNT ERRLOWOUTPUT ERRLOWOUTPUT HRBP HRBP ERRLOWOUTPUT->HRBP HR HR HR->HRSAT HR->HRBP HREKG HREKG HR->HREKG CO CO HR->CO VENTLUNG->MINVOL VENTLUNG->VENTALV EXPCO2 EXPCO2 VENTLUNG->EXPCO2 ARTCO2 ARTCO2 CATECHOL CATECHOL ARTCO2->CATECHOL ARTCO2->EXPCO2 VENTTUBE VENTTUBE VENTMACH->VENTTUBE PVSAT PVSAT SAO2 SAO2 PVSAT->SAO2 PCWP PCWP LVEDVOLUME->PCWP CVP CVP LVEDVOLUME->CVP FIO2 FIO2 FIO2->PVSAT HISTORY HISTORY TPR->CATECHOL BP BP TPR->BP VENTALV->ARTCO2 VENTALV->PVSAT STROKEVOLUME->CO ERRCAUTER ERRCAUTER ERRCAUTER->HRSAT ERRCAUTER->HREKG LVFAILURE LVFAILURE LVFAILURE->LVEDVOLUME LVFAILURE->HISTORY LVFAILURE->STROKEVOLUME KINKEDTUBE KINKEDTUBE KINKEDTUBE->VENTLUNG KINKEDTUBE->PRESS DISCONNECT DISCONNECT DISCONNECT->VENTTUBE SHUNT->SAO2 CATECHOL->HR VENTTUBE->VENTLUNG VENTTUBE->PRESS INSUFFANESTH INSUFFANESTH INSUFFANESTH->CATECHOL SAO2->CATECHOL CO->BP
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 : -152262.307411605
out/train.csv: 100%|███████████████████████████████████████|
Log2-Likelihood : -30278.502447798623

Out[3]:
-30278.502447798623

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 PULMEMBOLUS PULMEMBOLUS SHUNT SHUNT PULMEMBOLUS->SHUNT CATECHOL CATECHOL PULMEMBOLUS->CATECHOL ANAPHYLAXIS ANAPHYLAXIS TPR TPR ANAPHYLAXIS->TPR PAP PAP PAP->PULMEMBOLUS MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH HYPOVOLEMIA HYPOVOLEMIA STROKEVOLUME STROKEVOLUME HYPOVOLEMIA->STROKEVOLUME HRSAT HRSAT HREKG HREKG HRSAT->HREKG HRBP HRBP MINVOL MINVOL ERRLOWOUTPUT ERRLOWOUTPUT ERRLOWOUTPUT->HRBP HR HR HR->HRSAT HR->HRBP HR->HREKG CO CO HR->CO INTUBATION INTUBATION PRESS PRESS INTUBATION->PRESS INTUBATION->SHUNT VENTLUNG VENTLUNG VENTLUNG->MINVOL VENTALV VENTALV VENTLUNG->VENTALV KINKEDTUBE KINKEDTUBE VENTLUNG->KINKEDTUBE EXPCO2 EXPCO2 VENTLUNG->EXPCO2 ARTCO2 ARTCO2 VENTTUBE VENTTUBE VENTMACH->VENTTUBE PVSAT PVSAT SAO2 SAO2 PVSAT->SAO2 LVEDVOLUME LVEDVOLUME LVEDVOLUME->HYPOVOLEMIA LVFAILURE LVFAILURE LVEDVOLUME->LVFAILURE CVP CVP LVEDVOLUME->CVP FIO2 FIO2 FIO2->PVSAT HISTORY HISTORY PRESS->KINKEDTUBE VENTALV->INTUBATION VENTALV->ARTCO2 VENTALV->PVSAT STROKEVOLUME->CO BP BP TPR->BP PCWP PCWP PCWP->LVEDVOLUME ERRCAUTER ERRCAUTER ERRCAUTER->HRSAT ERRCAUTER->HREKG LVFAILURE->HISTORY LVFAILURE->STROKEVOLUME DISCONNECT DISCONNECT DISCONNECT->VENTTUBE SHUNT->SAO2 CATECHOL->HR VENTTUBE->VENTLUNG VENTTUBE->PRESS INSUFFANESTH INSUFFANESTH SAO2->CATECHOL CO->BP
Learned with 10000 lines in 0.255s
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 PRESS 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 MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH PULMEMBOLUS PULMEMBOLUS PAP PAP PULMEMBOLUS->SHUNT CATECHOL CATECHOL PULMEMBOLUS->CATECHOL INSUFFANESTH INSUFFANESTH 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 HISTORY HISTORY LVFAILURE->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 LVEDVOLUME->HYPOVOLEMIA LVEDVOLUME->LVFAILURE PCWP PCWP CVP CVP LVEDVOLUME->CVP VENTMACH->VENTTUBE PCWP->LVEDVOLUME SAO2 SAO2 SHUNT->SAO2 VENTTUBE->VENTLUNG VENTTUBE->PRESS VENTLUNG->KINKEDTUBE VENTLUNG->MINVOL VENTLUNG->VENTALV EXPCO2 EXPCO2 VENTLUNG->EXPCO2 PRESS->KINKEDTUBE VENTALV->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 HYPOVOLEMIA->LVEDVOLUME INTUBATION INTUBATION SHUNT SHUNT INTUBATION->SHUNT MINVOL MINVOL INTUBATION->MINVOL INTUBATION->PRESS VENTALV VENTALV MINVOLSET MINVOLSET VENTMACH VENTMACH MINVOLSET->VENTMACH PULMEMBOLUS PULMEMBOLUS PULMEMBOLUS->KINKEDTUBE PAP PAP PULMEMBOLUS->PAP PULMEMBOLUS->SHUNT VENTTUBE VENTTUBE PULMEMBOLUS->VENTTUBE INSUFFANESTH INSUFFANESTH CATECHOL CATECHOL INSUFFANESTH->CATECHOL ERRLOWOUTPUT ERRLOWOUTPUT HISTORY HISTORY ERRLOWOUTPUT->HISTORY HRBP HRBP ERRCAUTER ERRCAUTER HRSAT HRSAT HREKG HREKG FIO2 FIO2 PVSAT PVSAT FIO2->PVSAT LVFAILURE LVFAILURE LVFAILURE->HISTORY DISCONNECT DISCONNECT DISCONNECT->INTUBATION DISCONNECT->VENTTUBE ANAPHYLAXIS ANAPHYLAXIS TPR TPR STROKEVOLUME->LVFAILURE STROKEVOLUME->LVEDVOLUME HR HR STROKEVOLUME->HR CO CO STROKEVOLUME->CO TPR->ANAPHYLAXIS BP BP TPR->BP LVEDVOLUME->LVFAILURE PCWP PCWP LVEDVOLUME->PCWP CVP CVP LVEDVOLUME->CVP VENTMACH->VENTTUBE ARTCO2 ARTCO2 SHUNT->ARTCO2 SAO2 SAO2 SHUNT->SAO2 HISTORY->PULMEMBOLUS VENTTUBE->KINKEDTUBE VENTTUBE->INTUBATION VENTTUBE->VENTLUNG VENTTUBE->PRESS EXPCO2 EXPCO2 VENTTUBE->EXPCO2 VENTLUNG->KINKEDTUBE VENTLUNG->INTUBATION VENTLUNG->MINVOL VENTLUNG->VENTALV VENTLUNG->ARTCO2 VENTALV->INTUBATION VENTALV->ARTCO2 VENTALV->PVSAT ARTCO2->ERRCAUTER ARTCO2->CATECHOL PVSAT->SAO2 EXPCO2->VENTLUNG EXPCO2->VENTALV EXPCO2->ARTCO2 CATECHOL->PULMEMBOLUS CATECHOL->TPR CATECHOL->SAO2 HR->ERRLOWOUTPUT HR->ERRCAUTER HR->CATECHOL HR->HRSAT HR->HREKG HRBP->ERRLOWOUTPUT HRBP->HR HRBP->CO HRSAT->ERRCAUTER CO->HR 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 VENTTUBE VENTTUBE PULMEMBOLUS->VENTTUBE INSUFFANESTH INSUFFANESTH CATECHOL CATECHOL INSUFFANESTH->CATECHOL ERRLOWOUTPUT ERRLOWOUTPUT HISTORY HISTORY ERRLOWOUTPUT->HISTORY 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 LVFAILURE->HISTORY DISCONNECT DISCONNECT DISCONNECT->INTUBATION DISCONNECT->VENTTUBE ANAPHYLAXIS ANAPHYLAXIS ANAPHYLAXIS->ERRLOWOUTPUT TPR TPR ANAPHYLAXIS->TPR 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->PULMEMBOLUS 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())
{'PULMEMBOLUS', 'ANAPHYLAXIS', 'PAP', 'MINVOLSET', 'HYPOVOLEMIA', 'HRSAT', 'HRBP', 'MINVOL', 'ERRLOWOUTPUT', 'HR', 'INTUBATION', 'VENTLUNG', 'ARTCO2', 'VENTMACH', 'PVSAT', 'LVEDVOLUME', 'FIO2', 'HISTORY', 'PRESS', 'HREKG', 'VENTALV', 'STROKEVOLUME', 'TPR', 'PCWP', 'ERRCAUTER', 'LVFAILURE', 'DISCONNECT', 'KINKEDTUBE', 'SHUNT', 'CATECHOL', 'CVP', 'EXPCO2', 'VENTTUBE', 'INSUFFANESTH', 'SAO2', 'BP', 'CO'}
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