33 KiB
33 KiB
<html lang="en">
<head>
</head>
</html>
| LCOV - code coverage report | ||||||||||||||||||||||
 | ||||||||||||||||||||||
|
||||||||||||||||||||||
|
Line data Source code 1 : // *************************************************************** 2 : // SPDX-FileCopyrightText: Copyright 2024 Ricardo Montañana Gómez 3 : // SPDX-FileType: SOURCE 4 : // SPDX-License-Identifier: MIT 5 : // *************************************************************** 6 : 7 : #include "Ensemble.h" 8 : 9 : namespace bayesnet { 10 : 11 156 : Ensemble::Ensemble(bool predict_voting) : Classifier(Network()), n_models(0), predict_voting(predict_voting) 12 : { 13 : 14 156 : }; 15 : const std::string ENSEMBLE_NOT_FITTED = "Ensemble has not been fitted"; 16 12 : void Ensemble::trainModel(const torch::Tensor& weights) 17 : { 18 12 : n_models = models.size(); 19 94 : for (auto i = 0; i < n_models; ++i) { 20 : // fit with std::vectors 21 82 : models[i]->fit(dataset, features, className, states); 22 : } 23 12 : } 24 22 : std::vector<int> Ensemble::compute_arg_max(std::vector<std::vector<double>>& X) 25 : { 26 22 : std::vector<int> y_pred; 27 4910 : for (auto i = 0; i < X.size(); ++i) { 28 4888 : auto max = std::max_element(X[i].begin(), X[i].end()); 29 9776 : y_pred.push_back(std::distance(X[i].begin(), max)); 30 : } 31 22 : return y_pred; 32 0 : } 33 212 : torch::Tensor Ensemble::compute_arg_max(torch::Tensor& X) 34 : { 35 212 : auto y_pred = torch::argmax(X, 1); 36 212 : return y_pred; 37 : } 38 80 : torch::Tensor Ensemble::voting(torch::Tensor& votes) 39 : { 40 : // Convert m x n_models tensor to a m x n_class_states with voting probabilities 41 80 : auto y_pred_ = votes.accessor<int, 2>(); 42 80 : std::vector<int> y_pred_final; 43 80 : int numClasses = states.at(className).size(); 44 : // votes is m x n_models with the prediction of every model for each sample 45 80 : auto result = torch::zeros({ votes.size(0), numClasses }, torch::kFloat32); 46 80 : auto sum = std::reduce(significanceModels.begin(), significanceModels.end()); 47 20612 : for (int i = 0; i < votes.size(0); ++i) { 48 : // n_votes store in each index (value of class) the significance added by each model 49 : // i.e. n_votes[0] contains how much value has the value 0 of class. That value is generated by the models predictions 50 20532 : std::vector<double> n_votes(numClasses, 0.0); 51 171800 : for (int j = 0; j < n_models; ++j) { 52 151268 : n_votes[y_pred_[i][j]] += significanceModels.at(j); 53 : } 54 20532 : result[i] = torch::tensor(n_votes); 55 20532 : } 56 : // To only do one division and gain precision 57 80 : result /= sum; 58 160 : return result; 59 80 : } 60 44 : std::vector<std::vector<double>> Ensemble::predict_proba(std::vector<std::vector<int>>& X) 61 : { 62 44 : if (!fitted) { 63 12 : throw std::logic_error(ENSEMBLE_NOT_FITTED); 64 : } 65 32 : return predict_voting ? predict_average_voting(X) : predict_average_proba(X); 66 : } 67 226 : torch::Tensor Ensemble::predict_proba(torch::Tensor& X) 68 : { 69 226 : if (!fitted) { 70 12 : throw std::logic_error(ENSEMBLE_NOT_FITTED); 71 : } 72 214 : return predict_voting ? predict_average_voting(X) : predict_average_proba(X); 73 : } 74 28 : std::vector<int> Ensemble::predict(std::vector<std::vector<int>>& X) 75 : { 76 28 : auto res = predict_proba(X); 77 40 : return compute_arg_max(res); 78 20 : } 79 218 : torch::Tensor Ensemble::predict(torch::Tensor& X) 80 : { 81 218 : auto res = predict_proba(X); 82 420 : return compute_arg_max(res); 83 210 : } 84 148 : torch::Tensor Ensemble::predict_average_proba(torch::Tensor& X) 85 : { 86 148 : auto n_states = models[0]->getClassNumStates(); 87 148 : torch::Tensor y_pred = torch::zeros({ X.size(1), n_states }, torch::kFloat32); 88 148 : auto threads{ std::vector<std::thread>() }; 89 148 : std::mutex mtx; 90 882 : for (auto i = 0; i < n_models; ++i) { 91 734 : threads.push_back(std::thread([&, i]() { 92 734 : auto ypredict = models[i]->predict_proba(X); 93 734 : std::lock_guard<std::mutex> lock(mtx); 94 734 : y_pred += ypredict * significanceModels[i]; 95 734 : })); 96 : } 97 882 : for (auto& thread : threads) { 98 734 : thread.join(); 99 : } 100 148 : auto sum = std::reduce(significanceModels.begin(), significanceModels.end()); 101 148 : y_pred /= sum; 102 296 : return y_pred; 103 148 : } 104 18 : std::vector<std::vector<double>> Ensemble::predict_average_proba(std::vector<std::vector<int>>& X) 105 : { 106 18 : auto n_states = models[0]->getClassNumStates(); 107 18 : std::vector<std::vector<double>> y_pred(X[0].size(), std::vector<double>(n_states, 0.0)); 108 18 : auto threads{ std::vector<std::thread>() }; 109 18 : std::mutex mtx; 110 140 : for (auto i = 0; i < n_models; ++i) { 111 122 : threads.push_back(std::thread([&, i]() { 112 122 : auto ypredict = models[i]->predict_proba(X); 113 122 : assert(ypredict.size() == y_pred.size()); 114 122 : assert(ypredict[0].size() == y_pred[0].size()); 115 122 : std::lock_guard<std::mutex> lock(mtx); 116 : // Multiply each prediction by the significance of the model and then add it to the final prediction 117 24182 : for (auto j = 0; j < ypredict.size(); ++j) { 118 24060 : std::transform(y_pred[j].begin(), y_pred[j].end(), ypredict[j].begin(), y_pred[j].begin(), 119 154020 : [significanceModels = significanceModels[i]](double x, double y) { return x + y * significanceModels; }); 120 : } 121 122 : })); 122 : } 123 140 : for (auto& thread : threads) { 124 122 : thread.join(); 125 : } 126 18 : auto sum = std::reduce(significanceModels.begin(), significanceModels.end()); 127 : //Divide each element of the prediction by the sum of the significances 128 3358 : for (auto j = 0; j < y_pred.size(); ++j) { 129 19780 : std::transform(y_pred[j].begin(), y_pred[j].end(), y_pred[j].begin(), [sum](double x) { return x / sum; }); 130 : } 131 36 : return y_pred; 132 18 : } 133 14 : std::vector<std::vector<double>> Ensemble::predict_average_voting(std::vector<std::vector<int>>& X) 134 : { 135 14 : torch::Tensor Xt = bayesnet::vectorToTensor(X, false); 136 14 : auto y_pred = predict_average_voting(Xt); 137 14 : std::vector<std::vector<double>> result = tensorToVectorDouble(y_pred); 138 28 : return result; 139 14 : } 140 80 : torch::Tensor Ensemble::predict_average_voting(torch::Tensor& X) 141 : { 142 : // Build a m x n_models tensor with the predictions of each model 143 80 : torch::Tensor y_pred = torch::zeros({ X.size(1), n_models }, torch::kInt32); 144 80 : auto threads{ std::vector<std::thread>() }; 145 80 : std::mutex mtx; 146 616 : for (auto i = 0; i < n_models; ++i) { 147 536 : threads.push_back(std::thread([&, i]() { 148 536 : auto ypredict = models[i]->predict(X); 149 536 : std::lock_guard<std::mutex> lock(mtx); 150 1608 : y_pred.index_put_({ "...", i }, ypredict); 151 1072 : })); 152 : } 153 616 : for (auto& thread : threads) { 154 536 : thread.join(); 155 : } 156 160 : return voting(y_pred); 157 80 : } 158 40 : float Ensemble::score(torch::Tensor& X, torch::Tensor& y) 159 : { 160 40 : auto y_pred = predict(X); 161 36 : int correct = 0; 162 11292 : for (int i = 0; i < y_pred.size(0); ++i) { 163 11256 : if (y_pred[i].item<int>() == y[i].item<int>()) { 164 9834 : correct++; 165 : } 166 : } 167 72 : return (double)correct / y_pred.size(0); 168 36 : } 169 20 : float Ensemble::score(std::vector<std::vector<int>>& X, std::vector<int>& y) 170 : { 171 20 : auto y_pred = predict(X); 172 16 : int correct = 0; 173 4292 : for (int i = 0; i < y_pred.size(); ++i) { 174 4276 : if (y_pred[i] == y[i]) { 175 3574 : correct++; 176 : } 177 : } 178 32 : return (double)correct / y_pred.size(); 179 16 : } 180 2 : std::vector<std::string> Ensemble::show() const 181 : { 182 2 : auto result = std::vector<std::string>(); 183 10 : for (auto i = 0; i < n_models; ++i) { 184 8 : auto res = models[i]->show(); 185 8 : result.insert(result.end(), res.begin(), res.end()); 186 8 : } 187 2 : return result; 188 0 : } 189 6 : std::vector<std::string> Ensemble::graph(const std::string& title) const 190 : { 191 6 : auto result = std::vector<std::string>(); 192 40 : for (auto i = 0; i < n_models; ++i) { 193 34 : auto res = models[i]->graph(title + "_" + std::to_string(i)); 194 34 : result.insert(result.end(), res.begin(), res.end()); 195 34 : } 196 6 : return result; 197 0 : } 198 12 : int Ensemble::getNumberOfNodes() const 199 : { 200 12 : int nodes = 0; 201 100 : for (auto i = 0; i < n_models; ++i) { 202 88 : nodes += models[i]->getNumberOfNodes(); 203 : } 204 12 : return nodes; 205 : } 206 12 : int Ensemble::getNumberOfEdges() const 207 : { 208 12 : int edges = 0; 209 100 : for (auto i = 0; i < n_models; ++i) { 210 88 : edges += models[i]->getNumberOfEdges(); 211 : } 212 12 : return edges; 213 : } 214 2 : int Ensemble::getNumberOfStates() const 215 : { 216 2 : int nstates = 0; 217 10 : for (auto i = 0; i < n_models; ++i) { 218 8 : nstates += models[i]->getNumberOfStates(); 219 : } 220 2 : return nstates; 221 : } 222 : } |
|
| Generated by: LCOV version 2.0-1 |
</html>