-
Notifications
You must be signed in to change notification settings - Fork 2
/
example_inference.cpp
79 lines (63 loc) · 3.41 KB
/
example_inference.cpp
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
#include <baylib/smile_utils/smile_utils.hpp>
#include <baylib/inference/gibbs_sampling.hpp>
#include "baylib/inference/opencl/logic_sampling_opencl.hpp"
#include <baylib/inference/likelihood_weighting.hpp>
#include <iostream>
// Evidences in a bayesian network are a way to include posteriori knowledge
// about the state of the network and analyze how this knowledge affects the
// rest of the network
int main(int argc, char** argv){
using namespace baylib;
using namespace baylib::inference;
// Evidences can be applied to all types of networks, we use Credit as an example
// https://repo.bayesfusion.com/network/permalink?net=Small+BNs%2FCredit.xdsl
xdsl_parser<double> parser;
auto network = parser.deserialize("../../examples/xdsl/Credit.xdsl");
auto name_map = make_name_map(network);
// Here we declare the Gibbs sampling algorithm to perform approximate inference
// PLEASE NOTICE: Gibbs Sampling fails with bayesian networks with deterministic nodes
// (it computes wrong marginal probabilities, it's a well known theoretical
// limit of this sampling approach) hence it should not be used in such cases
gibbs_sampling gibbs_sampler(network, 10000, 4);
// You can obtain the inference output calling the make_inference
// method. Every baylib's inference algorithm has it.
// The output is a global marginal distribution for the network such that
// marginal[i][j] == P(i = j)
// where i is the id of the random_variable
// j is the id of the state
auto inf_result = gibbs_sampler.make_inference();
std::cout << gibbs_sampler.make_inference();
// you can also prettify the output if your network contains
// named nodes
for(ulong i = 0; i < network.number_of_variables(); ++i)
for(ulong j = 0; j < inf_result[i].size(); ++j)
std::cout << "P(" << network[i].name() << "=" << network[i].state(j) << ") = "
<< inf_result[i][j] << '\n';
std::cout << "\n\n";
// Let's now assume we know, as evidence, the values of "Debit" and "Income"
// to see how to inference simulation behaves
unsigned long Debit = name_map["Debit"];
unsigned long Income = name_map["Income"];
network[Debit].set_as_evidence(0);
network[Income].set_as_evidence(1);
// To detect if a node was set as evidence you can use the is_evidence and evidence_state methods
if(network[Debit].is_evidence())
std::cout << network[Debit].evidence_state() << '\n';
// The algorithm will automatically detect all evidences set and use them in the inferences
std::cout << gibbs_sampler.make_inference() << '\n';
// To clear the evidences use the clear_evidence on the desired nodes or use the util clear_network_evidences
// to clean all the network from evidences
network[Debit].clear_evidence();
clear_network_evidences(network);
// The network now should be on its base state
// Let's know try with a different algorithm to see if the results are coherent
// with gibbs sampling: likelihood weighting
likelihood_weighting likely_weigh(network, 10000, 4);
std::cout << likely_weigh.make_inference() << '\n';
// Different evidences can now be specified
// let's see how likelihood-weighting behaves
network[Debit].set_as_evidence(1);
network[Income].set_as_evidence(0);
std::cout << likely_weigh.make_inference() << '\n';
return 0;
}