# Some examples of do-calculus¶

In [1]:

from IPython.display import display, Math, Latex, HTML

import pyAgrum as gum
import pyAgrum.lib.notebook as gnb
import pyAgrum.causal as csl
import pyAgrum.causal.notebook as cslnb


## S. Tikka and J. Karvanen, 2016 [CRAN]¶

In [2]:

bn = gum.fastBN("w->x->z->y;w->z")

bn.cpt("w")[:] = [0.7,0.3]

bn.cpt("x")[:] = [[0.4,0.6],[0.3,0.7]]

bn.cpt("z")[{'w':0,'x':0}]=[0.2,0.8]
bn.cpt("z")[{'w':0,'x':1}]=[0.1,0.9]
bn.cpt("z")[{'w':1,'x':0}]=[0.9,0.1]
bn.cpt("z")[{'w':1,'x':1}]=[0.5,0.5]

bn.cpt("y")[:] = [[0.1,0.9],[0.8,0.2]]

d = csl.CausalModel(bn, [("lat1", ["x","y"])])
#csl.causalImpact(d,"y",{"x":0})
cslnb.showCausalImpact(d,"y","x",values={"x":0})
cslnb.showCausalImpact(d,"y","x",values={"x":1})


Causal Model
$$\begin{equation*}P( y \mid \hookrightarrow\mkern-6.5mux) = \sum_{w,z}{\left(\sum_{x'}{P\left(y\mid w,x',z\right) \cdot P\left(x'\mid w\right)}\right) \cdot P\left(z\mid w,x\right) \cdot P\left(w\right)}\end{equation*}$$
Explanation : Do-calculus computations
y
0
1
0.51300.4870

Impact

Causal Model
$$\begin{equation*}P( y \mid \hookrightarrow\mkern-6.5mux) = \sum_{w,z}{\left(\sum_{x'}{P\left(y\mid w,x',z\right) \cdot P\left(x'\mid w\right)}\right) \cdot P\left(z\mid w,x\right) \cdot P\left(w\right)}\end{equation*}$$
Explanation : Do-calculus computations
y
0
1
0.64600.3540

Impact

Since we have the formula, let us compute by hand this intervention :

In [3]:

(((bn.cpt("x") * bn.cpt("y")).margSumOut(["x"]) * bn.cpt("w") * bn.cpt("z")).margSumOut(["z", "w"])).putFirst("y")

Out[3]:

y
x
0
1
0
0.51300.4870
1
0.64600.3540
In [4]:

bn = gum.fastBN("Z1->X->Z2->Y")
d=csl.CausalModel(bn,[("L1",["Z1","X"]),
("L2",["Z1","Z2"]),
("L3",["Z1","Y"]),
("L4",["Y","X"])],
True)
cslnb.showCausalImpact(d,"Y","X",values={"X":1})


Causal Model
Hedge Error: G={'Z2', 'Z1', 'X'}, G[S]={'Z2'}
Impossible
No result
Impact

## Front door¶

In [5]:

modele4 = gum.BayesNet()

#Smoking
modele4.cpt(0)[:]=[0.5, 0.5]

#Tar
modele4.cpt(2)[{"Smoking":0}] = [0.4, 0.6]
modele4.cpt(2)[{"Smoking":1}] = [0.3, 0.6]

#Cancer
modele4.cpt(1)[{'Smoking':0,'Tar':0}]=[0.1,0.9] #No Drug, Male -> healed in 0.8 of cases
modele4.cpt(1)[{'Smoking':0,'Tar':1}]=[0.15,0.85] #No Drug, Female -> healed in 0.4 of cases
modele4.cpt(1)[{'Smoking':1,'Tar':0}]=[0.2,0.8] #Drug, Male -> healed 0.7 of cases
modele4.cpt(1)[{'Smoking':1,'Tar':1}]=[0.25,0.75]

d4 = csl.CausalModel(modele4, [("Genotype", ["Smoking","Cancer"])],False)
cslnb.showCausalModel(d4)

In [6]:

try:
a = csl.doCalculusWithObservation (d4,"Cancer", {"Smoking"})
except csl.HedgeException as h:
print (h.message)

In [7]:

display(Math(a.toLatex()))

$\displaystyle P( Cancer \mid \hookrightarrow\mkern-6.5muSmoking) = \sum_{Tar}{P\left(Tar\mid Smoking\right) \cdot \left(\sum_{Smoking'}{P\left(Smoking'\right) \cdot P\left(Cancer\mid Smoking',Tar\right)}\right)}$
In [8]:

try:
except csl.UnidentifiableException as u:
print (u.message)



||  Cancer           |
Smokin||0        |1        |
------||---------|---------|
0     || 0.1774  | 0.8226  |
1     || 0.1626  | 0.7374  |


In [9]:

formula, adj, exp = csl.causalImpact(d4, "Cancer", "Smoking",values={"Smoking":0})

In [10]:

display(Math(formula.toLatex()))

$\displaystyle P( Cancer \mid \hookrightarrow\mkern-6.5muSmoking) = \sum_{Tar}{P\left(Tar\mid Smoking\right) \cdot \left(\sum_{Smoking'}{P\left(Cancer\mid Smoking',Tar\right) \cdot P\left(Smoking'\right)}\right)}$
Out[10]:

Cancer
0
1
0.17740.8226

## Last example from R

In [11]:

m = gum.fastBN("z2->x->z1->y;z2->z1;z2->z3->y")

m.cpt("z2") [:] = [0.5, 0.5]
m.cpt("x") [:] = [[0.4,0.6], #z2=0
[0.4,0.6]] #z2=1
m.cpt("z3") [:] = [[0.3,0.7], #z2=0
[0.3,0.7]] #z2=1
m.cpt("z1")[{"z2":0, "x":0}]= [0.2, 0.8]
m.cpt("z1")[{"z2":0, "x":1}]= [0.25, 0.75]
m.cpt("z1")[{"z2":1, "x":0}]= [0.1, 0.9]
m.cpt("z1")[{"z2":1, "x":1}]= [0.15, 0.85]

m.cpt("y")[{"z1":0,"z3":0}]= [0.5,0.5]
m.cpt("y")[{"z1":0,"z3":1}]= [0.45,0.55]
m.cpt("y")[{"z1":1,"z3":0}]= [0.4,0.6]
m.cpt("y")[{"z1":1,"z3":1}]= [0.35,0.65]

d = csl.CausalModel(m, [("X-Z2",["x","z2"]),
("X-Z3",["x","z3"]),
("X-Y",["x","y"]),
("Y-Z2",["y","z2"])],
True)

cslnb.showCausalModel(d)

In [12]:

try:
formula,result,msg = csl.causalImpact(d,on={"y","z2","z1","z3"},doing={"x"})
except csl.HedgeException as h:
print (h.message)

print(msg)
display(Math(formula.toLatex()))

Do-calculus computations

$\displaystyle P( z1,y,z3,z2 \mid \hookrightarrow\mkern-6.5mux) = P\left(z3\mid z2\right) \cdot P\left(z1\mid x,z2\right) \cdot \frac {\sum_{x'}{P\left(y\mid x',z1,z2,z3\right) \cdot P\left(x'\mid z2\right) \cdot P\left(z3\mid x',z2\right) \cdot P\left(z2\right)}}{\sum_{x',y'}{P\left(y'\mid x',z1,z2,z3\right) \cdot P\left(x'\mid z2\right) \cdot P\left(z3\mid x',z2\right) \cdot P\left(z2\right)}} \cdot P\left(z2\right)$
In [13]:

# computation for this formula directly in pyAgrum
f1=m.cpt("x")*m.cpt("z2")*m.cpt("z3")*m.cpt("y")
f2=f1.margSumOut(["x"])
f3=f1.margSumOut(["x","y"])
f4=f2/f3
pyResult=m.cpt("z3")*m.cpt("z1")*m.cpt("z2")*f4

In [14]:

# computation for this formula directly by creating the causal AST
a = csl.ASTposteriorProba(m,{"z1"},{"x","z2"})
b= csl.ASTposteriorProba(m,{"y","z3"},{"x","z1","z2"})
c = csl.ASTjointProba(["x","z2"])
correct = csl.ASTmult(a,csl.ASTsum(["x"],csl.ASTmult(b,c)))

print("According to [ref], the result should be :")
display(Math(correct.toLatex()))

According to [ref], the result should be :

$\displaystyle P\left(z1\mid x,z2\right) \cdot \left(\sum_{x}{P\left(y,z3\mid z1,z2\right) \cdot P\left(x,z2\right)}\right)$
In [15]:

# computation for that formula
ie=gum.LazyPropagation(m)
refResult=(ie.evidenceJointImpact(["y","z3"],["x","z1","z2"])*
ie.evidenceJointImpact(["x","z2"],[])
).margSumOut(["x"])* m.cpt("z1")

In [16]:

print("Maximum error between these 3 versions : {}".format(max((refResult-pyResult).abs().max(),
(refResult-result).abs().max(),
(pyResult-result).new_abs().max())))

Maximum error between these 3 versions : 5.551115123125783e-17


## Unidentifiabilty¶

In [17]:

m1 = gum.fastBN("z1->x->z2->y")

cdg = csl.CausalModel(m1, [("Z1−X",["z1","x"]),
("Z1-Y",["z1","y"]),
("Z1-Z1",["z1","z2"]),
("X−Y",["x","y"])
], True )
cslnb.showCausalModel(cdg)

In [18]:

err = cslnb.showCausalImpact(cdg,"y","x",values={"x":0})


Causal Model
Hedge Error: G={'z1', 'x', 'z2'}, G[S]={'z2'}
Impossible
No result
Impact

## another one¶

In [19]:

# EXEMPLE PAGE 17 : http://ftp.cs.ucla.edu/pub/stat_ser/r350.pdf

m1 = gum.BayesNet()

gnb.showBN(m1)
d = csl.CausalModel(m1)

In [20]:

display(Math(csl.identifyingIntervention (d,{"z1","z2","z3","y"}, {"x"}).toLatex()))

$\displaystyle P\left(z3\mid z1,z2\right) \cdot P\left(z2\right) \cdot P\left(z1\right) \cdot P\left(y\mid x,z2,z3\right)$
In [21]:

display(Math(csl.identifyingIntervention(d,{"y"}, {"x"}).toLatex()))

$\displaystyle \sum_{z1,z2,z3}{P\left(z3\mid z1,z2\right) \cdot P\left(z2\right) \cdot P\left(z1\right) \cdot P\left(y\mid x,z2,z3\right)}$
In [22]:

display(Math(csl.identifyingIntervention(d,{"z1","z3","y"}, {"x","z2"}).toLatex()))

$\displaystyle P\left(z3\mid z1,z2\right) \cdot P\left(z1\right) \cdot P\left(y\mid x,z2,z3\right)$
In [23]:

display(Math(csl.identifyingIntervention(d,{"y"}, {"x","z2"}).toLatex()))

$\displaystyle \sum_{z1,z3}{P\left(z3\mid z1,z2\right) \cdot P\left(z1\right) \cdot P\left(y\mid x,z2,z3\right)}$

## Other example¶

In [24]:

#http://www.stats.ox.ac.uk/~lienart/gml15_causalinference.html

m1 = gum.BayesNet()

gnb.showBN(m1)
d = csl.CausalModel(m1)

In [25]:

display(Math(csl.identifyingIntervention(d,{"y"}, {"a","b"}).toLatex()))

$\displaystyle \sum_{p}{P\left(y\mid a,b,p\right) \cdot P\left(p\mid a\right)}$

## example f¶

In [26]:

#https://cse.sc.edu/~mgv/talks/AIM2010.ppt , example (f)

m1 = gum.BayesNet()

d = csl.CausalModel(m1, [("l1",["X","Z1"]) ,("l2",["Y","Z1"])],True)
cslnb.showCausalModel(d)

In [27]:

try:
display(Math(csl.identifyingIntervention (d,{"Y"}, {"X"}).toLatex()))
except csl.HedgeException as e:
print("Hedge exception : {}".format(e))

Hedge exception : Hedge Error: G={'Y', 'Z1', 'X'}, G[S]={'Z1', 'Y'}


## Example [Pearl,2009] Causality, p66¶

In [28]:

bn = gum.fastBN("Z1->Z2->Z3->Y<-X->Z2;Z2->Y;Z1->X->Z3<-Z1")
gnb.showBN(bn)

In [29]:

c = csl.CausalModel(bn, [("Z0", ("X", "Z1", "Z3"))], False)
cslnb.showCausalModel(c)

In [30]:

formula, impact, explanation = csl.causalImpact(c, "Y", "X")
cslnb.showCausalImpact(c,"Y","X")


Causal Model
$$\begin{equation*}P( Y \mid \hookrightarrow\mkern-6.5muX) = \sum_{Z1,Z2,Z3}{P\left(Y\mid X,Z2,Z3\right) \cdot P\left(Z2\mid X,Z1\right) \cdot \left(\sum_{X'}{P\left(Z3\mid X',Z1,Z2\right) \cdot P\left(Z1\right) \cdot P\left(X'\mid Z1\right)}\right)}\end{equation*}$$
Explanation : Do-calculus computations
Y
X
0
1
0
0.46150.5385
1
0.42080.5792

Impact
In [ ]: