Different sampling inference

In [1]:

import os

%matplotlib inline
from pylab import *
import matplotlib.pyplot as plt

In [2]:

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

#gnb.showBN(bn,size='8')
print(bn)

BN{nodes: 413, arcs: 602, domainSize: 10^406.108, dim: 429409, mem: 3Mo 530Ko 104o}


First, some helpers

In [3]:

import timeit

class Timer:
def __enter__(self):
self.start =  timeit.default_timer()
return self

def __exit__(self, *args):
self.end = timeit.default_timer()
self.duration = self.end - self.start

In [4]:

def execute(ie):
with Timer() as t:
ie.makeInference()
for i in bn.nodes():
a=ie.posterior(i)
return "duration : {:3.3f}s".format(t.duration)

def vals(bn,ie):
exact=[]
appro=[]
for node in bn.nodes():
# potentials as numpy array
exact+=ie.posterior(node).tolist()

return exact


Exact inference.

In [5]:

import matplotlib_inline
matplotlib_inline.backend_inline.set_matplotlib_formats('png')

plt.rcParams["figure.figsize"] = [30,3]

def compareIE(bn,maxtime,epsilon,evs=None):
ie=gum.LazyPropagation(bn)
if evs is not None:
ie.setEvidence(evs)
x=vals(bn,ie)

ie2=gum.GibbsSampling(bn)
if evs is not None:
ie2.setEvidence(evs)
ie2.setMaxTime(maxtime)
ie2.setEpsilon(epsilon)
txt="Gibbs : "+execute(ie2)+"\n"+ie2.messageApproximationScheme()
y=vals(bn,ie2)
plt.subplot(181)
plt.plot(x,y,'ro')
plt.title(txt)

ie3=gum.MonteCarloSampling(bn)
if evs is not None:
ie3.setEvidence(evs)
ie3.setMaxTime(maxtime)
ie3.setEpsilon(epsilon)
txt="MonteCarlo : "+execute(ie3)+"\n"+ie3.messageApproximationScheme()
y=vals(bn,ie3)
plt.subplot(182)
plt.plot(x,y,'ro')
plt.title(txt)

ie4=gum.WeightedSampling(bn)
if evs is not None:
ie4.setEvidence(evs)
ie4.setMaxTime(maxtime)
ie4.setEpsilon(epsilon)
txt="Weighted : "+execute(ie4)+"\n"+ie4.messageApproximationScheme()
y=vals(bn,ie4)
plt.subplot(183)
plt.plot(x,y,'ro')
plt.title(txt)

ie5=gum.ImportanceSampling(bn)
if evs is not None:
ie5.setEvidence(evs)
ie5.setMaxTime(maxtime)
ie5.setEpsilon(epsilon)
txt="Importance: "+execute(ie5)+"\n"+ie5.messageApproximationScheme()
y=vals(bn,ie5)
plt.subplot(184)
plt.plot(x,y,'ro')
plt.title(txt)

ie6=gum.LoopyBeliefPropagation(bn)
if evs is not None:
ie6.setEvidence(evs)
ie6.setMaxTime(maxtime)
ie6.setEpsilon(epsilon)
txt="LBP: "+execute(ie6)+"\n"+ie6.messageApproximationScheme()
y=vals(bn,ie6)
plt.subplot(185)
plt.plot(x,y,'ro')
plt.title(txt)

ie7=gum.LoopyWeightedSampling(bn)
if evs is not None:
ie7.setEvidence(evs)
ie7.setMaxTime(maxtime)
ie7.setEpsilon(epsilon)
txt="LoopyWeighted: "+execute(ie7)+"\n"+ie7.messageApproximationScheme()
y=vals(bn,ie7)
plt.subplot(186)
plt.plot(x,y,'ro')
plt.title(txt)

ie8=gum.LoopyGibbsSampling(bn)
if evs is not None:
ie8.setEvidence(evs)
ie8.setMaxTime(maxtime)
ie8.setEpsilon(epsilon)
txt="LoopyGibbs: "+execute(ie8)+"\n"+ie8.messageApproximationScheme()
y=vals(bn,ie8)
plt.subplot(187)
plt.plot(x,y,'ro')
plt.title(txt)

ie9=gum.LoopyImportanceSampling(bn)
if evs is not None:
ie9.setEvidence(evs)
ie9.setMaxTime(maxtime)
ie9.setEpsilon(epsilon)
txt="LoopyImportance: "+execute(ie9)+"\n"+ie9.messageApproximationScheme()
y=vals(bn,ie9)
plt.subplot(188)
plt.plot(x,y,'ro')
plt.title(txt)

plt.show()

In [6]:

compareIE(bn,5,1e-2)

In [7]:

compareIE(bn,50,1e-2)

IOStream.flush timed out

In [8]:

compareIE(bn,5,1e-2,evs={'bg_24':0,'ins_indep_util_23':1,'renal_cl_14':1})

In [9]:

compareIE(bn,100,1e-2,evs={'bg_24':0,'ins_indep_util_23':1,'renal_cl_14':1})

IOStream.flush timed out