| 54 | !*/ |
| 55 | |
| 56 | struct regression_tree |
| 57 | { |
| 58 | std::vector<split_feature> splits; |
| 59 | std::vector<matrix<float,0,1> > leaf_values; |
| 60 | |
| 61 | unsigned long num_leaves() const { return leaf_values.size(); } |
| 62 | |
| 63 | inline const matrix<float,0,1>& operator()( |
| 64 | const std::vector<float>& feature_pixel_values, |
| 65 | unsigned long& i |
| 66 | ) const |
| 67 | /*! |
| 68 | requires |
| 69 | - All the index values in splits are less than feature_pixel_values.size() |
| 70 | - leaf_values.size() is a power of 2. |
| 71 | (i.e. we require a tree with all the levels fully filled out. |
| 72 | - leaf_values.size() == splits.size()+1 |
| 73 | (i.e. there needs to be the right number of leaves given the number of splits in the tree) |
| 74 | ensures |
| 75 | - runs through the tree and returns the vector at the leaf we end up in. |
| 76 | - #i == the selected leaf node index. |
| 77 | !*/ |
| 78 | { |
| 79 | i = 0; |
| 80 | while (i < splits.size()) |
| 81 | { |
| 82 | if ((float)feature_pixel_values[splits[i].idx1] - (float)feature_pixel_values[splits[i].idx2] > splits[i].thresh) |
| 83 | i = left_child(i); |
| 84 | else |
| 85 | i = right_child(i); |
| 86 | } |
| 87 | i = i - splits.size(); |
| 88 | return leaf_values[i]; |
| 89 | } |
| 90 | |
| 91 | friend void serialize (const regression_tree& item, std::ostream& out) |
| 92 | { |
| 93 | dlib::serialize(item.splits, out); |
| 94 | dlib::serialize(item.leaf_values, out); |
| 95 | } |
| 96 | friend void deserialize (regression_tree& item, std::istream& in) |
| 97 | { |
| 98 | dlib::deserialize(item.splits, in); |
| 99 | dlib::deserialize(item.leaf_values, in); |
| 100 | } |
| 101 | }; |
| 102 | |
| 103 | // ------------------------------------------------------------------------------------ |
| 104 |
nothing calls this directly
no test coverage detected