Practical - 4
AIM :- Burglary Alarm Example Using Bayesian
Network
Importing Library
In [29]:
1 from pgmpy.models import BayesianModel
2 from pgmpy.inference import VariableElimination
3 import bnlearn
Defining network structure
In [30]:
1 from pgmpy.factors.discrete import TabularCPD
2
3 alarm_model = BayesianModel([('Burglary', 'Alarm'), ('Earthquake', 'Alarm'),
4 ('Alarm', 'JohnCalls'), ('Alarm', 'MaryCalls')])
5
6 # Defining the parameters using CPT
7
8 cpd_burglary = TabularCPD(variable='Burglary',
9 variable_card=2,
10 values=[[.999], [0.001]])
11 cpd_earthquake = TabularCPD(variable='Earthquake',
12 variable_card=2,
13 values=[[0.998], [0.002]])
14 cpd_alarm = TabularCPD(variable='Alarm',
15 variable_card=2,
16 values=[[0.999, 0.71, 0.06, 0.05],
17 [0.001, 0.29, 0.94, 0.95]],
18 evidence=['Burglary', 'Earthquake'],
19 evidence_card=[2, 2])
20 cpd_johncalls = TabularCPD(variable='JohnCalls',
21 variable_card=2,
22 values=[[0.95, 0.1], [0.05, 0.9]],
23 evidence=['Alarm'],
24 evidence_card=[2])
25 cpd_marycalls = TabularCPD(variable='MaryCalls',
26 variable_card=2,
27 values=[[0.1, 0.7], [0.9, 0.3]],
28 evidence=['Alarm'],
29 evidence_card=[2])
30
31 # Associating the parameters with the model structure
32 alarm_model.add_cpds(cpd_burglary, cpd_earthquake, cpd_alarm, cpd_johncalls,
33 cpd_marycalls)
Checking if the cpds are valid for the model
In [31]:
1 alarm_model.check_model()
Out[31]:
True
Viewing nodes of the model
In [32]:
1 alarm_model.nodes()
Out[32]:
NodeView(('Burglary', 'Alarm', 'Earthquake', 'JohnCalls', 'MaryCalls'))
Viewing edges of the model
In [33]:
1 alarm_model.edges()
Out[33]:
OutEdgeView([('Burglary', 'Alarm'), ('Alarm', 'JohnCalls'), ('Alarm', 'MaryC
alls'), ('Earthquake', 'Alarm')])
Checking independcies of a node
In [34]:
1 alarm_model.local_independencies('Burglary')
Out[34]:
(Burglary ⟂ Earthquake)
Listing all Independencies
In [35]:
1 alarm_model.get_independencies()
Out[35]:
(MaryCalls ⟂ Earthquake, JohnCalls, Burglary | Alarm)
(MaryCalls ⟂ JohnCalls, Burglary | Earthquake, Alarm)
(MaryCalls ⟂ Earthquake, Burglary | JohnCalls, Alarm)
(MaryCalls ⟂ Earthquake, JohnCalls | Burglary, Alarm)
(MaryCalls ⟂ Burglary | Earthquake, JohnCalls, Alarm)
(MaryCalls ⟂ JohnCalls | Earthquake, Burglary, Alarm)
(MaryCalls ⟂ Earthquake | Burglary, JohnCalls, Alarm)
(Earthquake ⟂ Burglary)
(Earthquake ⟂ MaryCalls, JohnCalls | Alarm)
(Earthquake ⟂ JohnCalls | MaryCalls, Alarm)
(Earthquake ⟂ MaryCalls | JohnCalls, Alarm)
(Earthquake ⟂ MaryCalls, JohnCalls | Burglary, Alarm)
(Earthquake ⟂ JohnCalls | MaryCalls, Burglary, Alarm)
(Earthquake ⟂ MaryCalls | Burglary, JohnCalls, Alarm)
(JohnCalls ⟂ Earthquake, MaryCalls, Burglary | Alarm)
(JohnCalls ⟂ MaryCalls, Burglary | Earthquake, Alarm)
(JohnCalls ⟂ Earthquake, Burglary | MaryCalls, Alarm)
(JohnCalls ⟂ Earthquake, MaryCalls | Burglary, Alarm)
In [36]:
1 infer = VariableElimination(alarm_model)
Probability Distribution that are Explicitly Spelled out in our Graphs
In [37]:
1 alarm_prob = infer.query(['Alarm'])
2 print(alarm_prob)
0%| | 0/2 [00:00<?, ?it/s]
0%| | 0/2 [00:00<?, ?it/s]
+----------+--------------+
| Alarm | phi(Alarm) |
+==========+==============+
| Alarm(0) | 0.9975 |
+----------+--------------+
| Alarm(1) | 0.0025 |
+----------+--------------+
Conditional Probability of Ringing of Alarm When Burglary and Earthquake
is not occuring
In [38]:
1 print(infer.query(['Alarm'], evidence={'Burglary': 0, 'Earthquake': 0}))
0it [00:00, ?it/s]
0it [00:00, ?it/s]
+----------+--------------+
| Alarm | phi(Alarm) |
+==========+==============+
| Alarm(0) | 0.9990 |
+----------+--------------+
| Alarm(1) | 0.0010 |
+----------+--------------+
Conditional Probability Of Ringing Of Alarm When There Is No Burglary And
Earthquake But Still John And Merry Call Harry
In [39]:
1 print(
2 infer.query(['Alarm'],
3 evidence={
4 'Burglary': 0,
5 'Earthquake': 0,
6 'JohnCalls': 1,
7 'MaryCalls': 1
8 }))
0it [00:00, ?it/s]
0it [00:00, ?it/s]
+----------+--------------+
| Alarm | phi(Alarm) |
+==========+==============+
| Alarm(0) | 0.9940 |
+----------+--------------+
| Alarm(1) | 0.0060 |
+----------+--------------+
Conditional Probability of Ringing of Alarm When There is Burglary
Happens and Only Marry Calls Harry
In [40]:
1 print(
2 infer.query(['Alarm'],
3 evidence={
4 'Burglary': 1,
5 'Earthquake': 0,
6 'JohnCalls': 0,
7 'MaryCalls': 1
8 }))
0it [00:00, ?it/s]
0it [00:00, ?it/s]
+----------+--------------+
| Alarm | phi(Alarm) |
+==========+==============+
| Alarm(0) | 0.6453 |
+----------+--------------+
| Alarm(1) | 0.3547 |
+----------+--------------+
Conditional Probability of Ringing of Alarm When There is Burglary,
Earthquake and Both Call Harry
In [41]:
1 print(
2 infer.query(['Alarm'],
3 evidence={
4 'Burglary': 1,
5 'Earthquake': 1,
6 'JohnCalls': 1,
7 'MaryCalls': 1
8 }))
0it [00:00, ?it/s]
0it [00:00, ?it/s]
+----------+--------------+
| Alarm | phi(Alarm) |
+==========+==============+
| Alarm(0) | 0.0087 |
+----------+--------------+
| Alarm(1) | 0.9913 |
+----------+--------------+
