| 59 | typename labels_type |
| 60 | > |
| 61 | void make_training_examples( |
| 62 | dlib::array<graph_type>& samples, |
| 63 | labels_type& labels |
| 64 | ) |
| 65 | { |
| 66 | /* |
| 67 | This function makes 3 graphs we will use for training. All of them |
| 68 | will contain 4 nodes and have the structure shown below: |
| 69 | |
| 70 | (0)-----(1) |
| 71 | | | |
| 72 | | | |
| 73 | | | |
| 74 | (3)-----(2) |
| 75 | |
| 76 | In this example, each node has a 2-D vector. The first element of this vector |
| 77 | is 1 when the node should have a label of false while the second element has |
| 78 | a value of 1 when the node should have a label of true. Additionally, the |
| 79 | edge vectors will contain a value of 1 when the nodes connected by the edge |
| 80 | should share the same label and a value of 0 otherwise. |
| 81 | |
| 82 | We want to see that the machine learning method is able to figure out how |
| 83 | these features relate to the labels. If it is successful it will create a |
| 84 | graph_labeler which can predict the correct labels for these and other |
| 85 | similarly constructed graphs. |
| 86 | |
| 87 | Finally, note that these tools require all values in the edge vectors to be >= 0. |
| 88 | However, the node vectors may contain both positive and negative values. |
| 89 | */ |
| 90 | |
| 91 | samples.clear(); |
| 92 | labels.clear(); |
| 93 | |
| 94 | std::vector<bool> label; |
| 95 | graph_type g; |
| 96 | |
| 97 | // --------------------------- |
| 98 | g.set_number_of_nodes(4); |
| 99 | label.resize(g.number_of_nodes()); |
| 100 | // store the vector [0,1] into node 0. Also label it as true. |
| 101 | g.node(0).data = 0, 1; label[0] = true; |
| 102 | // store the vector [0,0] into node 1. |
| 103 | g.node(1).data = 0, 0; label[1] = true; // Note that this node's vector doesn't tell us how to label it. |
| 104 | // We need to take the edges into account to get it right. |
| 105 | // store the vector [1,0] into node 2. |
| 106 | g.node(2).data = 1, 0; label[2] = false; |
| 107 | // store the vector [0,0] into node 3. |
| 108 | g.node(3).data = 0, 0; label[3] = false; |
| 109 | |
| 110 | // Add the 4 edges as shown in the ASCII art above. |
| 111 | g.add_edge(0,1); |
| 112 | g.add_edge(1,2); |
| 113 | g.add_edge(2,3); |
| 114 | g.add_edge(3,0); |
| 115 | |
| 116 | // set the 1-D vector for the edge between node 0 and 1 to the value of 1. |
| 117 | edge(g,0,1) = 1; |
| 118 | // set the 1-D vector for the edge between node 1 and 2 to the value of 0. |
no test coverage detected