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.

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]:

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 : -152270.78324279713

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

Log2-Likelihood : -30047.566310365473

Out[3]:

-30047.566310365473

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()

Learned with 10000 lines in 0.485s

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()

Diff with MIIC

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()

Diff with GHC/NMD/BDEU

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()

Diff with GHC/NMD/BDEU with intial DAG from MIIC

In [9]:

print(bn2.names())

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

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