| 1 | /* ---------------------------------------------------------------------------- |
| 2 | |
| 3 | * GTSAM Copyright 2010, Georgia Tech Research Corporation, |
| 4 | * Atlanta, Georgia 30332-0415 |
| 5 | * All Rights Reserved |
| 6 | * Authors: Frank Dellaert, et al. (see THANKS for the full author list) |
| 7 | |
| 8 | * See LICENSE for the license information |
| 9 | |
| 10 | * -------------------------------------------------------------------------- */ |
| 11 | |
| 12 | /** |
| 13 | * @file DiscreteBayesNet_FG.cpp |
| 14 | * @brief Discrete Bayes Net example using Factor Graphs |
| 15 | * @author Abhijit |
| 16 | * @date Jun 4, 2012 |
| 17 | * |
| 18 | * We use the famous Rain/Cloudy/Sprinkler Example of [Russell & Norvig, 2009, |
| 19 | * p529] You may be familiar with other graphical model packages like BNT |
| 20 | * (available at http://bnt.googlecode.com/svn/trunk/docs/usage.html) where this |
| 21 | * is used as an example. The following demo is same as that in the above link, |
| 22 | * except that everything is using GTSAM. |
| 23 | */ |
| 24 | |
| 25 | #include <gtsam/discrete/DiscreteFactorGraph.h> |
| 26 | #include <gtsam/discrete/DiscreteMarginals.h> |
| 27 | |
| 28 | #include <iomanip> |
| 29 | |
| 30 | using namespace std; |
| 31 | using namespace gtsam; |
| 32 | |
| 33 | int main(int argc, char **argv) { |
| 34 | // Define keys and a print function |
| 35 | Key C(1), S(2), R(3), W(4); |
| 36 | auto print = [=](const DiscreteFactor::Values& values) { |
| 37 | cout << boolalpha << "Cloudy = " << static_cast<bool>(values.at(k: C)) |
| 38 | << " Sprinkler = " << static_cast<bool>(values.at(k: S)) |
| 39 | << " Rain = " << boolalpha << static_cast<bool>(values.at(k: R)) |
| 40 | << " WetGrass = " << static_cast<bool>(values.at(k: W)) << endl; |
| 41 | }; |
| 42 | |
| 43 | // We assume binary state variables |
| 44 | // we have 0 == "False" and 1 == "True" |
| 45 | const size_t nrStates = 2; |
| 46 | |
| 47 | // define variables |
| 48 | DiscreteKey Cloudy(C, nrStates), Sprinkler(S, nrStates), Rain(R, nrStates), |
| 49 | WetGrass(W, nrStates); |
| 50 | |
| 51 | // create Factor Graph of the bayes net |
| 52 | DiscreteFactorGraph graph; |
| 53 | |
| 54 | // add factors |
| 55 | graph.add(args&: Cloudy, args: "0.5 0.5" ); // P(Cloudy) |
| 56 | graph.add(args: Cloudy & Sprinkler, args: "0.5 0.5 0.9 0.1" ); // P(Sprinkler | Cloudy) |
| 57 | graph.add(args: Cloudy & Rain, args: "0.8 0.2 0.2 0.8" ); // P(Rain | Cloudy) |
| 58 | graph.add(args&: Sprinkler & Rain & WetGrass, |
| 59 | args: "1 0 0.1 0.9 0.1 0.9 0.001 0.99" ); // P(WetGrass | Sprinkler, Rain) |
| 60 | |
| 61 | // Alternatively we can also create a DiscreteBayesNet, add |
| 62 | // DiscreteConditional factors and create a FactorGraph from it. (See |
| 63 | // testDiscreteBayesNet.cpp) |
| 64 | |
| 65 | // Since this is a relatively small distribution, we can as well print |
| 66 | // the whole distribution.. |
| 67 | cout << "Distribution of Example: " << endl; |
| 68 | cout << setw(11) << "Cloudy(C)" << setw(14) << "Sprinkler(S)" << setw(10) |
| 69 | << "Rain(R)" << setw(14) << "WetGrass(W)" << setw(15) << "P(C,S,R,W)" |
| 70 | << endl; |
| 71 | for (size_t a = 0; a < nrStates; a++) |
| 72 | for (size_t m = 0; m < nrStates; m++) |
| 73 | for (size_t h = 0; h < nrStates; h++) |
| 74 | for (size_t c = 0; c < nrStates; c++) { |
| 75 | DiscreteFactor::Values values; |
| 76 | values[C] = c; |
| 77 | values[S] = h; |
| 78 | values[R] = m; |
| 79 | values[W] = a; |
| 80 | double prodPot = graph(values); |
| 81 | cout << setw(8) << static_cast<bool>(c) << setw(14) |
| 82 | << static_cast<bool>(h) << setw(12) << static_cast<bool>(m) |
| 83 | << setw(13) << static_cast<bool>(a) << setw(16) << prodPot |
| 84 | << endl; |
| 85 | } |
| 86 | |
| 87 | // "Most Probable Explanation", i.e., configuration with largest value |
| 88 | auto mpe = graph.optimize(); |
| 89 | cout << "\nMost Probable Explanation (MPE):" << endl; |
| 90 | print(mpe); |
| 91 | |
| 92 | // "Inference" We show an inference query like: probability that the Sprinkler |
| 93 | // was on; given that the grass is wet i.e. P( S | C=0) = ? |
| 94 | |
| 95 | // add evidence that it is not Cloudy |
| 96 | graph.add(args&: Cloudy, args: "1 0" ); |
| 97 | |
| 98 | // solve again, now with evidence |
| 99 | auto mpe_with_evidence = graph.optimize(); |
| 100 | |
| 101 | cout << "\nMPE given C=0:" << endl; |
| 102 | print(mpe_with_evidence); |
| 103 | |
| 104 | // we can also calculate arbitrary marginals: |
| 105 | DiscreteMarginals marginals(graph); |
| 106 | cout << "\nP(S=1|C=0):" << marginals.marginalProbabilities(key: Sprinkler)[1] |
| 107 | << endl; |
| 108 | cout << "\nP(R=0|C=0):" << marginals.marginalProbabilities(key: Rain)[0] << endl; |
| 109 | cout << "\nP(W=1|C=0):" << marginals.marginalProbabilities(key: WetGrass)[1] |
| 110 | << endl; |
| 111 | |
| 112 | // We can also sample from the eliminated graph |
| 113 | DiscreteBayesNet::shared_ptr chordal = graph.eliminateSequential(); |
| 114 | cout << "\n10 samples:" << endl; |
| 115 | for (size_t i = 0; i < 10; i++) { |
| 116 | auto sample = chordal->sample(); |
| 117 | print(sample); |
| 118 | } |
| 119 | return 0; |
| 120 | } |
| 121 | |