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 <thread>
8 : #include <mutex>
9 : #include <sstream>
10 : #include "Network.h"
11 : #include "bayesnet/utils/bayesnetUtils.h"
12 : namespace bayesnet {
13 930 : Network::Network() : fitted{ false }, maxThreads{ 0.95 }, classNumStates{ 0 }, laplaceSmoothing{ 0 }
14 : {
15 930 : }
16 4 : Network::Network(float maxT) : fitted{ false }, maxThreads{ maxT }, classNumStates{ 0 }, laplaceSmoothing{ 0 }
17 : {
18 :
19 4 : }
20 888 : Network::Network(const Network& other) : laplaceSmoothing(other.laplaceSmoothing), features(other.features), className(other.className), classNumStates(other.getClassNumStates()),
21 1776 : maxThreads(other.getMaxThreads()), fitted(other.fitted), samples(other.samples)
22 : {
23 888 : if (samples.defined())
24 2 : samples = samples.clone();
25 898 : for (const auto& node : other.nodes) {
26 10 : nodes[node.first] = std::make_unique<Node>(*node.second);
27 : }
28 888 : }
29 634 : void Network::initialize()
30 : {
31 634 : features.clear();
32 634 : className = "";
33 634 : classNumStates = 0;
34 634 : fitted = false;
35 634 : nodes.clear();
36 634 : samples = torch::Tensor();
37 634 : }
38 894 : float Network::getMaxThreads() const
39 : {
40 894 : return maxThreads;
41 : }
42 24 : torch::Tensor& Network::getSamples()
43 : {
44 24 : return samples;
45 : }
46 13374 : void Network::addNode(const std::string& name)
47 : {
48 13374 : if (name == "") {
49 4 : throw std::invalid_argument("Node name cannot be empty");
50 : }
51 13370 : if (nodes.find(name) != nodes.end()) {
52 0 : return;
53 : }
54 13370 : if (find(features.begin(), features.end(), name) == features.end()) {
55 13370 : features.push_back(name);
56 : }
57 13370 : nodes[name] = std::make_unique<Node>(name);
58 : }
59 118 : std::vector<std::string> Network::getFeatures() const
60 : {
61 118 : return features;
62 : }
63 1070 : int Network::getClassNumStates() const
64 : {
65 1070 : return classNumStates;
66 : }
67 24 : int Network::getStates() const
68 : {
69 24 : int result = 0;
70 144 : for (auto& node : nodes) {
71 120 : result += node.second->getNumStates();
72 : }
73 24 : return result;
74 : }
75 1590160 : std::string Network::getClassName() const
76 : {
77 1590160 : return className;
78 : }
79 30532 : bool Network::isCyclic(const std::string& nodeId, std::unordered_set<std::string>& visited, std::unordered_set<std::string>& recStack)
80 : {
81 30532 : if (visited.find(nodeId) == visited.end()) // if node hasn't been visited yet
82 : {
83 30532 : visited.insert(nodeId);
84 30532 : recStack.insert(nodeId);
85 36110 : for (Node* child : nodes[nodeId]->getChildren()) {
86 5590 : if (visited.find(child->getName()) == visited.end() && isCyclic(child->getName(), visited, recStack))
87 12 : return true;
88 5582 : if (recStack.find(child->getName()) != recStack.end())
89 4 : return true;
90 : }
91 : }
92 30520 : recStack.erase(nodeId); // remove node from recursion stack before function ends
93 30520 : return false;
94 : }
95 24954 : void Network::addEdge(const std::string& parent, const std::string& child)
96 : {
97 24954 : if (nodes.find(parent) == nodes.end()) {
98 4 : throw std::invalid_argument("Parent node " + parent + " does not exist");
99 : }
100 24950 : if (nodes.find(child) == nodes.end()) {
101 4 : throw std::invalid_argument("Child node " + child + " does not exist");
102 : }
103 : // Temporarily add edge to check for cycles
104 24946 : nodes[parent]->addChild(nodes[child].get());
105 24946 : nodes[child]->addParent(nodes[parent].get());
106 24946 : std::unordered_set<std::string> visited;
107 24946 : std::unordered_set<std::string> recStack;
108 24946 : if (isCyclic(nodes[child]->getName(), visited, recStack)) // if adding this edge forms a cycle
109 : {
110 : // remove problematic edge
111 4 : nodes[parent]->removeChild(nodes[child].get());
112 4 : nodes[child]->removeParent(nodes[parent].get());
113 4 : throw std::invalid_argument("Adding this edge forms a cycle in the graph.");
114 : }
115 24950 : }
116 1590294 : std::map<std::string, std::unique_ptr<Node>>& Network::getNodes()
117 : {
118 1590294 : return nodes;
119 : }
120 712 : void Network::checkFitData(int n_samples, int n_features, int n_samples_y, const std::vector<std::string>& featureNames, const std::string& className, const std::map<std::string, std::vector<int>>& states, const torch::Tensor& weights)
121 : {
122 712 : if (weights.size(0) != n_samples) {
123 4 : throw std::invalid_argument("Weights (" + std::to_string(weights.size(0)) + ") must have the same number of elements as samples (" + std::to_string(n_samples) + ") in Network::fit");
124 : }
125 708 : if (n_samples != n_samples_y) {
126 4 : throw std::invalid_argument("X and y must have the same number of samples in Network::fit (" + std::to_string(n_samples) + " != " + std::to_string(n_samples_y) + ")");
127 : }
128 704 : if (n_features != featureNames.size()) {
129 4 : throw std::invalid_argument("X and features must have the same number of features in Network::fit (" + std::to_string(n_features) + " != " + std::to_string(featureNames.size()) + ")");
130 : }
131 700 : if (features.size() == 0) {
132 4 : throw std::invalid_argument("The network has not been initialized. You must call addNode() before calling fit()");
133 : }
134 696 : if (n_features != features.size() - 1) {
135 4 : throw std::invalid_argument("X and local features must have the same number of features in Network::fit (" + std::to_string(n_features) + " != " + std::to_string(features.size() - 1) + ")");
136 : }
137 692 : if (find(features.begin(), features.end(), className) == features.end()) {
138 4 : throw std::invalid_argument("Class Name not found in Network::features");
139 : }
140 14210 : for (auto& feature : featureNames) {
141 13526 : if (find(features.begin(), features.end(), feature) == features.end()) {
142 4 : throw std::invalid_argument("Feature " + feature + " not found in Network::features");
143 : }
144 13522 : if (states.find(feature) == states.end()) {
145 0 : throw std::invalid_argument("Feature " + feature + " not found in states");
146 : }
147 : }
148 684 : }
149 684 : void Network::setStates(const std::map<std::string, std::vector<int>>& states)
150 : {
151 : // Set states to every Node in the network
152 684 : for_each(features.begin(), features.end(), [this, &states](const std::string& feature) {
153 14194 : nodes.at(feature)->setNumStates(states.at(feature).size());
154 14194 : });
155 684 : classNumStates = nodes.at(className)->getNumStates();
156 684 : }
157 : // X comes in nxm, where n is the number of features and m the number of samples
158 2 : void Network::fit(const torch::Tensor& X, const torch::Tensor& y, const torch::Tensor& weights, const std::vector<std::string>& featureNames, const std::string& className, const std::map<std::string, std::vector<int>>& states)
159 : {
160 2 : checkFitData(X.size(1), X.size(0), y.size(0), featureNames, className, states, weights);
161 2 : this->className = className;
162 2 : torch::Tensor ytmp = torch::transpose(y.view({ y.size(0), 1 }), 0, 1);
163 6 : samples = torch::cat({ X , ytmp }, 0);
164 10 : for (int i = 0; i < featureNames.size(); ++i) {
165 24 : auto row_feature = X.index({ i, "..." });
166 8 : }
167 2 : completeFit(states, weights);
168 12 : }
169 668 : void Network::fit(const torch::Tensor& samples, const torch::Tensor& weights, const std::vector<std::string>& featureNames, const std::string& className, const std::map<std::string, std::vector<int>>& states)
170 : {
171 668 : checkFitData(samples.size(1), samples.size(0) - 1, samples.size(1), featureNames, className, states, weights);
172 668 : this->className = className;
173 668 : this->samples = samples;
174 668 : completeFit(states, weights);
175 668 : }
176 : // input_data comes in nxm, where n is the number of features and m the number of samples
177 42 : void Network::fit(const std::vector<std::vector<int>>& input_data, const std::vector<int>& labels, const std::vector<double>& weights_, const std::vector<std::string>& featureNames, const std::string& className, const std::map<std::string, std::vector<int>>& states)
178 : {
179 42 : const torch::Tensor weights = torch::tensor(weights_, torch::kFloat64);
180 42 : checkFitData(input_data[0].size(), input_data.size(), labels.size(), featureNames, className, states, weights);
181 14 : this->className = className;
182 : // Build tensor of samples (nxm) (n+1 because of the class)
183 14 : samples = torch::zeros({ static_cast<int>(input_data.size() + 1), static_cast<int>(input_data[0].size()) }, torch::kInt32);
184 70 : for (int i = 0; i < featureNames.size(); ++i) {
185 224 : samples.index_put_({ i, "..." }, torch::tensor(input_data[i], torch::kInt32));
186 : }
187 56 : samples.index_put_({ -1, "..." }, torch::tensor(labels, torch::kInt32));
188 14 : completeFit(states, weights);
189 112 : }
190 684 : void Network::completeFit(const std::map<std::string, std::vector<int>>& states, const torch::Tensor& weights)
191 : {
192 684 : setStates(states);
193 684 : laplaceSmoothing = 1.0 / samples.size(1); // To use in CPT computation
194 684 : std::vector<std::thread> threads;
195 14878 : for (auto& node : nodes) {
196 14194 : threads.emplace_back([this, &node, &weights]() {
197 14194 : node.second->computeCPT(samples, features, laplaceSmoothing, weights);
198 14194 : });
199 : }
200 14878 : for (auto& thread : threads) {
201 14194 : thread.join();
202 : }
203 684 : fitted = true;
204 684 : }
205 1588 : torch::Tensor Network::predict_tensor(const torch::Tensor& samples, const bool proba)
206 : {
207 1588 : if (!fitted) {
208 4 : throw std::logic_error("You must call fit() before calling predict()");
209 : }
210 1584 : torch::Tensor result;
211 1584 : result = torch::zeros({ samples.size(1), classNumStates }, torch::kFloat64);
212 377028 : for (int i = 0; i < samples.size(1); ++i) {
213 1126344 : const torch::Tensor sample = samples.index({ "...", i });
214 375448 : auto psample = predict_sample(sample);
215 375444 : auto temp = torch::tensor(psample, torch::kFloat64);
216 : // result.index_put_({ i, "..." }, torch::tensor(predict_sample(sample), torch::kFloat64));
217 1126332 : result.index_put_({ i, "..." }, temp);
218 375448 : }
219 1580 : if (proba)
220 738 : return result;
221 1684 : return result.argmax(1);
222 752476 : }
223 : // Return mxn tensor of probabilities
224 738 : torch::Tensor Network::predict_proba(const torch::Tensor& samples)
225 : {
226 738 : return predict_tensor(samples, true);
227 : }
228 :
229 : // Return mxn tensor of probabilities
230 850 : torch::Tensor Network::predict(const torch::Tensor& samples)
231 : {
232 850 : return predict_tensor(samples, false);
233 : }
234 :
235 : // Return mx1 std::vector of predictions
236 : // tsamples is nxm std::vector of samples
237 24 : std::vector<int> Network::predict(const std::vector<std::vector<int>>& tsamples)
238 : {
239 24 : if (!fitted) {
240 8 : throw std::logic_error("You must call fit() before calling predict()");
241 : }
242 16 : std::vector<int> predictions;
243 16 : std::vector<int> sample;
244 1782 : for (int row = 0; row < tsamples[0].size(); ++row) {
245 1770 : sample.clear();
246 13126 : for (int col = 0; col < tsamples.size(); ++col) {
247 11356 : sample.push_back(tsamples[col][row]);
248 : }
249 1770 : std::vector<double> classProbabilities = predict_sample(sample);
250 : // Find the class with the maximum posterior probability
251 1766 : auto maxElem = max_element(classProbabilities.begin(), classProbabilities.end());
252 1766 : int predictedClass = distance(classProbabilities.begin(), maxElem);
253 1766 : predictions.push_back(predictedClass);
254 1766 : }
255 24 : return predictions;
256 20 : }
257 : // Return mxn std::vector of probabilities
258 : // tsamples is nxm std::vector of samples
259 132 : std::vector<std::vector<double>> Network::predict_proba(const std::vector<std::vector<int>>& tsamples)
260 : {
261 132 : if (!fitted) {
262 4 : throw std::logic_error("You must call fit() before calling predict_proba()");
263 : }
264 128 : std::vector<std::vector<double>> predictions;
265 128 : std::vector<int> sample;
266 24798 : for (int row = 0; row < tsamples[0].size(); ++row) {
267 24670 : sample.clear();
268 219650 : for (int col = 0; col < tsamples.size(); ++col) {
269 194980 : sample.push_back(tsamples[col][row]);
270 : }
271 24670 : predictions.push_back(predict_sample(sample));
272 : }
273 256 : return predictions;
274 128 : }
275 10 : double Network::score(const std::vector<std::vector<int>>& tsamples, const std::vector<int>& labels)
276 : {
277 10 : std::vector<int> y_pred = predict(tsamples);
278 6 : int correct = 0;
279 1162 : for (int i = 0; i < y_pred.size(); ++i) {
280 1156 : if (y_pred[i] == labels[i]) {
281 972 : correct++;
282 : }
283 : }
284 12 : return (double)correct / y_pred.size();
285 6 : }
286 : // Return 1xn std::vector of probabilities
287 26440 : std::vector<double> Network::predict_sample(const std::vector<int>& sample)
288 : {
289 : // Ensure the sample size is equal to the number of features
290 26440 : if (sample.size() != features.size() - 1) {
291 8 : throw std::invalid_argument("Sample size (" + std::to_string(sample.size()) +
292 12 : ") does not match the number of features (" + std::to_string(features.size() - 1) + ")");
293 : }
294 26436 : std::map<std::string, int> evidence;
295 232760 : for (int i = 0; i < sample.size(); ++i) {
296 206324 : evidence[features[i]] = sample[i];
297 : }
298 52872 : return exactInference(evidence);
299 26436 : }
300 : // Return 1xn std::vector of probabilities
301 375448 : std::vector<double> Network::predict_sample(const torch::Tensor& sample)
302 : {
303 : // Ensure the sample size is equal to the number of features
304 375448 : if (sample.size(0) != features.size() - 1) {
305 8 : throw std::invalid_argument("Sample size (" + std::to_string(sample.size(0)) +
306 12 : ") does not match the number of features (" + std::to_string(features.size() - 1) + ")");
307 : }
308 375444 : std::map<std::string, int> evidence;
309 8888488 : for (int i = 0; i < sample.size(0); ++i) {
310 8513044 : evidence[features[i]] = sample[i].item<int>();
311 : }
312 750888 : return exactInference(evidence);
313 375444 : }
314 1590148 : double Network::computeFactor(std::map<std::string, int>& completeEvidence)
315 : {
316 1590148 : double result = 1.0;
317 33392584 : for (auto& node : getNodes()) {
318 31802436 : result *= node.second->getFactorValue(completeEvidence);
319 : }
320 1590148 : return result;
321 : }
322 401880 : std::vector<double> Network::exactInference(std::map<std::string, int>& evidence)
323 : {
324 401880 : std::vector<double> result(classNumStates, 0.0);
325 401880 : std::vector<std::thread> threads;
326 401880 : std::mutex mtx;
327 1992028 : for (int i = 0; i < classNumStates; ++i) {
328 1590148 : threads.emplace_back([this, &result, &evidence, i, &mtx]() {
329 1590148 : auto completeEvidence = std::map<std::string, int>(evidence);
330 1590148 : completeEvidence[getClassName()] = i;
331 1590148 : double factor = computeFactor(completeEvidence);
332 1590148 : std::lock_guard<std::mutex> lock(mtx);
333 1590148 : result[i] = factor;
334 1590148 : });
335 : }
336 1992028 : for (auto& thread : threads) {
337 1590148 : thread.join();
338 : }
339 : // Normalize result
340 401880 : double sum = accumulate(result.begin(), result.end(), 0.0);
341 1992028 : transform(result.begin(), result.end(), result.begin(), [sum](const double& value) { return value / sum; });
342 803760 : return result;
343 401880 : }
344 14 : std::vector<std::string> Network::show() const
345 : {
346 14 : std::vector<std::string> result;
347 : // Draw the network
348 80 : for (auto& node : nodes) {
349 66 : std::string line = node.first + " -> ";
350 154 : for (auto child : node.second->getChildren()) {
351 88 : line += child->getName() + ", ";
352 : }
353 66 : result.push_back(line);
354 66 : }
355 14 : return result;
356 0 : }
357 44 : std::vector<std::string> Network::graph(const std::string& title) const
358 : {
359 44 : auto output = std::vector<std::string>();
360 44 : auto prefix = "digraph BayesNet {\nlabel=<BayesNet ";
361 44 : auto suffix = ">\nfontsize=30\nfontcolor=blue\nlabelloc=t\nlayout=circo\n";
362 44 : std::string header = prefix + title + suffix;
363 44 : output.push_back(header);
364 350 : for (auto& node : nodes) {
365 306 : auto result = node.second->graph(className);
366 306 : output.insert(output.end(), result.begin(), result.end());
367 306 : }
368 44 : output.push_back("}\n");
369 88 : return output;
370 44 : }
371 132 : std::vector<std::pair<std::string, std::string>> Network::getEdges() const
372 : {
373 132 : auto edges = std::vector<std::pair<std::string, std::string>>();
374 2906 : for (const auto& node : nodes) {
375 2774 : auto head = node.first;
376 7924 : for (const auto& child : node.second->getChildren()) {
377 5150 : auto tail = child->getName();
378 5150 : edges.push_back({ head, tail });
379 5150 : }
380 2774 : }
381 132 : return edges;
382 0 : }
383 110 : int Network::getNumEdges() const
384 : {
385 110 : return getEdges().size();
386 : }
387 110 : std::vector<std::string> Network::topological_sort()
388 : {
389 : /* Check if al the fathers of every node are before the node */
390 110 : auto result = features;
391 110 : result.erase(remove(result.begin(), result.end(), className), result.end());
392 110 : bool ending{ false };
393 314 : while (!ending) {
394 204 : ending = true;
395 1902 : for (auto feature : features) {
396 1698 : auto fathers = nodes[feature]->getParents();
397 4500 : for (const auto& father : fathers) {
398 2802 : auto fatherName = father->getName();
399 2802 : if (fatherName == className) {
400 1490 : continue;
401 : }
402 : // Check if father is placed before the actual feature
403 1312 : auto it = find(result.begin(), result.end(), fatherName);
404 1312 : if (it != result.end()) {
405 1312 : auto it2 = find(result.begin(), result.end(), feature);
406 1312 : if (it2 != result.end()) {
407 1312 : if (distance(it, it2) < 0) {
408 : // if it is not, insert it before the feature
409 122 : result.erase(remove(result.begin(), result.end(), fatherName), result.end());
410 122 : result.insert(it2, fatherName);
411 122 : ending = false;
412 : }
413 : }
414 : }
415 2802 : }
416 1698 : }
417 : }
418 110 : return result;
419 0 : }
420 4 : std::string Network::dump_cpt() const
421 : {
422 4 : std::stringstream oss;
423 24 : for (auto& node : nodes) {
424 20 : oss << "* " << node.first << ": (" << node.second->getNumStates() << ") : " << node.second->getCPT().sizes() << std::endl;
425 20 : oss << node.second->getCPT() << std::endl;
426 : }
427 8 : return oss.str();
428 4 : }
429 : }
|
