Influence diagram¶

| | |

In [1]:

import os

%matplotlib inline
from pylab import *
import matplotlib.pyplot as plt
from IPython.display import display,HTML

import math

In [2]:

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

Build a Influence Diagram¶

fast build with string¶

In [3]:

gum.fastID("A->*B->$C<-D<-*E->*G->H->*I<-D")

Out[3]:

bifxml format file¶

In [4]:

diag=gum.loadID("res/diag.bifxml")
gnb.showInfluenceDiagram(diag)

No description has been provided for this image

In [5]:

diag

Out[5]:

the hard way :-)¶

In [6]:

F=diag.addChanceNode(gum.LabelizedVariable("F","F",2))
diag.addArc(diag.idFromName("decisionVar1"),F)

U=diag.addUtilityNode(gum.LabelizedVariable("U","U",1))
diag.addArc(diag.idFromName("decisionVar3"),U)
diag.addArc(diag.idFromName("F"),U)
gnb.showInfluenceDiagram(diag)

In [7]:

diag.cpt(F)[{'decisionVar1':0}]=[0.9,0.1]
diag.cpt(F)[{'decisionVar1':1}]=[0.3,0.7]

diag.utility(U)[{'F':0,'decisionVar3':0}]=2
diag.utility(U)[{'F':0,'decisionVar3':1}]=4
diag.utility(U)[{'F':1}]=[[0],[5]]

Optimization in an influence diagram (actually LIMID)¶

In [8]:

oil=gum.loadID("res/OilWildcatter.bifxml")
gnb.flow.row(oil,gnb.getInference(oil))

In [9]:

# a function to show results on decision nodes T and D
def show_decisions(ie):
    gnb.flow.row(ie.optimalDecision("Testing"),
                   ie.optimalDecision("Drilling"),
                   f"$${ie.MEU()['mean']:5.3f}\\ (stdev : {math.sqrt(ie.MEU()['variance']):5.3f})$$",
                   captions=["Strategy for T",
                             "Strategy for D",
                             "MEU and its standard deviation"])
    gnb.flow.row(ie.posterior("Testing"),ie.posteriorUtility("Testing"),
                   ie.posterior("Drilling"),ie.posteriorUtility("Drilling"),
                  captions=["Final decision for Testing","Final reward for Testing",
                            "Final decision for Drilling","Final reward for Drilling"])

ie=gum.ShaferShenoyLIMIDInference(oil) 

display(HTML("<h2>Inference in the LIMID optimizing the decisions nodes</h2>"))
ie.makeInference()
show_decisions(ie)

Inference in the LIMID optimizing the decisions nodes

Testing
Yes	No
1.0000	0.0000

Strategy for T

	Drilling
TestResult	Yes	No
closed	1.0000	0.0000
open	1.0000	0.0000
diffuse	0.0000	1.0000

Strategy for D

$$22.500\ (stdev : 87.457)$$
MEU and its standard deviation

Testing
Yes	No
1.0000	0.0000

Final decision for Testing

Testing
Yes	No
22.5000	20.0000

Final reward for Testing

Drilling
Yes	No
0.5900	0.4100

Final decision for Drilling

Drilling
Yes	No
45.0847	-10.0000

Final reward for Drilling

Graphical inference with evidence and targets (developped nodes)¶

In [10]:

gnb.sideBySide(oil,
               gnb.getInference(oil,evs={'TestResult':'closed'}),
               gnb.getInference(oil,evs={'TestResult':'open'}),
               gnb.getInference(oil,evs={'TestResult':'diffuse'}),
               oil,
               gnb.getInference(oil,evs={'OilContents':'Dry'}),
               gnb.getInference(oil,evs={'OilContents':'Wet'}),
               gnb.getInference(oil,evs={'OilContents':'Soaking'}),
               ncols=4)

Soft evidence on chance node¶

In [11]:

gnb.showInference(oil,evs={'OilContents':[0.7,0.5,0.8]})

Forced decision¶

In [12]:

gnb.showInference(oil,evs={'Drilling':'Yes'})