-
Notifications
You must be signed in to change notification settings - Fork 44
/
README.txt
176 lines (163 loc) · 9.03 KB
/
README.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
libpgm is a set of tools to use probabilistic graphical models to compute
probabilities in multivariate systems.
To install in your default location for third-party packages, run:
[sudo] python setup.py install
Included in this package are:
1. Python modules containing PGM tools (libpgm/)
dictionary.py
discretebayesiannetwork.py
graphskeleton.py
hybayesiannetwork.py
lhbayesiannetwork.py
nodedata.py
orderedskeleton.py
pgmlearner.py
sampleaggregator.py
tablecpdfactorization.py
tablecpdfactor.py
/CPDtypes/discrete.py
/CPDtypes/crazy.py
/CPDtypes/lgandd.py
/CPDtypes/lg.py
2. Unit tests and unit test dictionaries to test functionality (tests/)
run_unit_tests.py
unittestdict.txt
unittestlgdict.txt
unittesthdict.txt
unittestdyndict.txt
3. Examples of how to use tools (/examples)
examples.py
exampleevidence.txt
examplequery.txt
exampledata.txt
4. Documentation (/docs/_build/_html)
index.html (has links to all documentation pages)
To use this library, one must create an input file or input files for their Bayesian network --
for examples of this, see the unit test dictionaries:
unittestdict.txt -- A sample input file for a discrete-CPD Bayesian
network
unittestlgdict.txt -- A sample input file for a linear Gaussian-CPD Bayesian
network
unittesthdict.txt -- A sample input file for a hybrid (any type) CPD
Bayesian network
unittestdyndict.txt -- A sample input file for a dynamic discrete Bayesian
network
Once the input file is created, this library allows for the user to execute a
number of queries and commands. These are listed below, each with a brief
example, a reference to Probabilistic Graphical Models by Koller and Friedman,
and a reference to an example in examples.py.
SAMPLING CAPABILITIES
1. Generate a sequence of samples from a discrete-CPD Bayesian network:
Load a "NodeData" instance and a "GraphSkeleton" instance from input
files. Run the "toporder" method on the GraphSkeleton instance to
topologically order the graph. Load a "DiscreteBayesianNetwork" instance
from the skeleton and node data. Run the "randomsample" method on the
DiscreteBayesianNetwork instance to return a sequence of samples.
Koller reference -- 12.1
Examples reference -- (1)
2. Generate a sequence of samples from a linear Gaussian-CPD Bayesian network:
Same as previous, but use "LGBayesianNetwork" instead of
"DiscreteBayesianNetwork".
Koller reference -- 12.1
Examples reference -- (2)
3. Generate a sequence of samples from a hybrid (any CPD type) Bayesian network:
For all of the CPD types, create a module, class, and "choose" method of the
type found in /CPDtypes/. This method must take in a list of parent outcomes
and return a properly sampled value for its node. Once this method is
written, load a "NodeData" instance and a "GraphSkeleton" instance from the input
file (and run "toporder" on the GraphSkeleton instance). Then run the
"entriestoinstances" method on the NodeData instance. Then load a
"HyBayesianNetwork" instance from the skeleton and node data. Then run
the "randomsample" method on the HyBayesianNetwork instance to
return a sequence of samples.
Koller reference -- 14.5.2
Examples reference -- (3)
4. Generate a sequence of samples from a discrete-CPD Bayesian network, given evidence:
Set up the Bayesian network as in (1). Place evidence in a .txt file in
the format found in exampleevidence.txt. Load a "TableCPDFactorization"
instance from the DiscreteBayesianNetwork instance. Run the "gibbssample" method
on the TableCPDFactorization instance which returns a sequence. These
samples provably get closer to the posterior probability distribution
(the probability distribution given evidence) over time.
Koller reference -- 12.3.1
Examples reference -- (4)
DETERMINISTIC INFERENCE CAPABILITIES (all for discrete-CPD Bayesian networks)
5. Compute the probability distribution over a specific node or nodes:
Set up a DiscreteBayesianNetwork instance as in (1). Load a
TableCPDFactorization instance from the DiscreteBayesianNetwork instance.
If you have evidence, place it in a dictionary in the style of
exampleevidence.txt. Place your query in a dictionary in the style of
examplequery.txt. Load both evidence and query into python dicts. Run
the "condprobve" method on the TableCPDFactorization instance, with
query and evidence dicts as arguments. This will return a factor (an
alternative form of a probability distribution -- see Koller 104) with
the probability distribution over the variables you queried about (it
will ignore the outcomes included in your query).
Koller reference -- 9.3
Examples reference -- (5)
6. Compute the exact probability of an outcome:
Set up a TableCPDFactorization instance, a query dict, and an
evidence dict in the manner specified by (5). Then run the
"specificquery" method on the TableCPDFactorization instance with query
and evidence as arguments. This will return the specific probability
that the outcomes specified by query will occur given the evidence
presented.
Koller reference -- 9.3
Examples reference -- (6)
APPROXIMATIVE INFERENCE CAPABILITIES
7. Compute the approximate probability distribution by generating samples:
Create an SampleAggregator instance. Instantiate a type of Bayesian
network class that you want to sample. Then run the "aggregate" method
with a sampling function statement as its argument (e.g.,
instance.randomsample(100)). This will return the "average" of all
samples generated, which is an approximate probability distribution.
Koller reference -- 12.1
Examples reference -- (7)
BAYESIAN NETWORK LEARNING CAPABILITIES
8. Learn the CPDs of a discrete Bayesian network, given data and a structure:
Load a GraphSkeleton instance from an inputfile that holds your graph structure.
Transform your data into an inputfile of the format found in
exampledata.txt. Then load a python dict from that input file
(using the eval function, or something else). Then create a
"PGMLearner" instance. Run the "discrete_mle_estimateparams" method with
the dict of data and the GraphSkeleton instance as arguments. This will return
a DiscreteBayesianNetwork instance with the CPDs filled out.
Koller reference -- 17.2.3
Examples reference -- (8)
9. Learn the structure of a discrete Bayesian network, given only data:
Load the data into a dict in the manner specified in (8). Load a
PGMLearner instance. Run the "discrete_constraint_estimatestruct" method
with the data as an argument (p-value threshhold and graph indegree
constraint are optional arguments as well). This will return a
GraphSkeleton instance learned from the data.
Koller reference -- 18.2
Examples reference -- (9)
10. Learn the CPDs of a linear Gaussian Bayesian network, given data and a structure:
Same as (8), but run the "lg_mle_estimateparams" method.
Koller reference -- 17.2.4
Examples reference -- (10)
11. Learn the structure of a linear Gaussian Bayesian network, given only data:
Same as (9), but run the "lg_constraint_estimatestruct". An additional
optional argument is the number of bins to use to discretize the continuous
data.
Koller reference -- 18.2
Examples reference -- (11)
12. Learn entire Bayesian networks (structures and parameters) from data:
Load the data and the PGMLearner instance as described in (8). Run
either "discrete_estimatebn" or "lg_estimatebn" depending on whether
your data is discrete or linear Gaussian. These methods simply combine
the functions above. They take all the optional arguments.
Koller reference -- 16-18
Examples reference -- (12)
DYNAMIC SAMPLING CAPABILITIES
13. Generate a sequence of samples over time from a dynamic discrete Bayesian network:
Load a "NodeData" and "GraphSkeleton" instance from proper input files. Run the
"toporder" method to topologically order the graph skeleton. Instantiate a
"DynDiscBayesianNetwork" instance from the NodeData and Graph Skeleton. Run the
method "randomsample" on the DynDiscBayesianNetwok to generate a sequence of samples,
one per time interval, with each influenced by the outcomes of its predecessor.
Koller reference -- 6.2
Examples reference -- (13)
--
CyberPoint International 2012
Created by Charlie Cabot