rmontanana/BayesNet
rmontanana/BayesNet
1
0
Fork 0
Code Issues 2 Pull Requests Packages Projects Releases 3 Wiki Activity

Begin with parameter estimation

Browse Source
This commit is contained in:
Ricardo Montañana Gómez 2023-06-30 21:24:12 +02:00
parent 0a31aa2ff1
commit 71d730d228
Signed by: rmontanana
GPG Key ID: 46064262FD9A7ADE
8 changed files with 236 additions and 87 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CMakeLists.txt
View File

@ -12,4 +12,6 @@ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${TORCH_CXX_FLAGS}")
# add_library(BayesNet Node.cc Network.cc)
add_executable(BayesNet main.cc ArffFiles.cc Node.cc Network.cc CPPFImdlp.cpp Metrics.cpp)
add_executable(test test.cc)
target_link_libraries(BayesNet "${TORCH_LIBRARIES}")
target_link_libraries(test "${TORCH_LIBRARIES}")

1
CPPFImdlp.cpp
View File

@ -37,6 +37,7 @@ namespace mdlp {
y = y_;
num_cut_points = compute_max_num_cut_points();
depth = 0;
discretizedData.clear();
cutPoints.clear();
if (X.size() != y.size()) {
throw invalid_argument("X and y must have the same size");

159
Network.cc
View File

@ -1,7 +1,7 @@
#include "Network.h"
namespace bayesnet {
Network::Network() : laplaceSmoothing(1), root(nullptr) {}
Network::Network(int smoothing) : laplaceSmoothing(smoothing), root(nullptr) {}
Network::Network() : laplaceSmoothing(1), root(nullptr), features(vector<string>()), className("") {}
Network::Network(int smoothing) : laplaceSmoothing(smoothing), root(nullptr), features(vector<string>()), className("") {}
Network::~Network()
{
for (auto& pair : nodes) {
@ -10,6 +10,9 @@ namespace bayesnet {
}
void Network::addNode(string name, int numStates)
{
if (nodes.find(name) != nodes.end()) {
throw invalid_argument("Node " + name + " already exists");
}
nodes[name] = new Node(name, numStates);
if (root == nullptr) {
root = nodes[name];
@ -32,7 +35,6 @@ namespace bayesnet {
{
visited.insert(nodeId);
recStack.insert(nodeId);
for (Node* child : nodes[nodeId]->getChildren()) {
if (visited.find(child->getName()) == visited.end() && isCyclic(child->getName(), visited, recStack))
return true;
@ -55,13 +57,11 @@ namespace bayesnet {
nodes[parent]->addChild(nodes[child]);
nodes[child]->addParent(nodes[parent]);
// temporarily add edge
unordered_set<string> visited;
unordered_set<string> recStack;
if (isCyclic(nodes[child]->getName(), visited, recStack)) // if adding this edge forms a cycle
{
// remove edge
// remove problematic edge
nodes[parent]->removeChild(nodes[child]);
nodes[child]->removeParent(nodes[parent]);
throw invalid_argument("Adding this edge forms a cycle in the graph.");
@ -72,40 +72,89 @@ namespace bayesnet {
{
return nodes;
}
void Network::buildNetwork(const vector<vector<int>>& dataset, const vector<int>& labels, const vector<string>& featureNames, const string& className)
void Network::buildNetwork()
{
// Add features as nodes to the network
for (int i = 0; i < featureNames.size(); ++i) {
addNode(featureNames[i], *max_element(dataset[i].begin(), dataset[i].end()) + 1);
for (int i = 0; i < features.size(); ++i) {
addNode(features[i], *max_element(dataset[features[i]].begin(), dataset[features[i]].end()) + 1);
}
// Add class as node to the network
addNode(className, *max_element(labels.begin(), labels.end()) + 1);
addNode(className, *max_element(dataset[className].begin(), dataset[className].end()) + 1);
// Add edges from class to features => naive Bayes
for (auto feature : featureNames) {
for (auto feature : features) {
addEdge(className, feature);
}
addEdge("petalwidth", "petallength");
}
void Network::fit(const vector<vector<int>>& dataset, const vector<int>& labels, const vector<string>& featureNames, const string& className)
{
buildNetwork(dataset, labels, featureNames, className);
//estimateParameters(dataset);
features = featureNames;
this->className = className;
// Build dataset
for (int i = 0; i < featureNames.size(); ++i) {
this->dataset[featureNames[i]] = dataset[i];
}
this->dataset[className] = labels;
buildNetwork();
estimateParameters();
}
// void Network::estimateParameters()
// {
// auto dimensions = vector<int64_t>();
// for (auto [name, node] : nodes) {
// // Get dimensions of the CPT
// dimensions.clear();
// dimensions.push_back(node->getNumStates());
// for (auto father : node->getParents()) {
// dimensions.push_back(father->getNumStates());
// }
// auto length = dimensions.size();
// // Create a tensor of zeros with the dimensions of the CPT
// torch::Tensor cpt = torch::zeros(dimensions, torch::kFloat);
// // Fill table with counts
// for (int n_sample = 0; n_sample < dataset[name].size(); ++n_sample) {
// torch::List<c10::optional<torch::Tensor>> coordinates;
// coordinates.push_back(torch::tensor(dataset[name][n_sample]));
// for (auto father : node->getParents()) {
// coordinates.push_back(torch::tensor(dataset[father->getName()][n_sample]));
// }
// // Increment the count of the corresponding coordinate
// cpt.index_put_({ coordinates }, cpt.index({ coordinates }) + 1);
// }
// // store thre resulting cpt in the node
// node->setCPT(cpt);
// }
// }
// void Network::estimateParameters()
// {
// // Lambda function to compute joint counts of states
// auto jointCounts = [this](const vector<string>& nodeNames) {
// int size = nodeNames.size();
// std::vector<int64_t> sizes(size);
// for (int i = 0; i < size; ++i) {
// sizes[i] = this->nodes[nodeNames[i]]->getNumStates();
// }
// auto jointCounts = [](const vector<vector<int>>& data, const vector<int>& indices, int numStates) {
// int size = indices.size();
// vector<int64_t> sizes(size, numStates);
// torch::Tensor counts = torch::zeros(sizes, torch::kLong);
// for (const auto& row : data) {
// int idx = 0;
// int dataSize = this->dataset[nodeNames[0]].size();
// for (int dataIdx = 0; dataIdx < dataSize; ++dataIdx) {
// std::vector<torch::Tensor> idx(size);
// for (int i = 0; i < size; ++i) {
// idx = idx * numStates + row[indices[i]];
// idx[i] = torch::tensor(this->dataset[nodeNames[i]][dataIdx], torch::kLong);
// }
// counts.view({ -1 }).add_(idx, 1);
// torch::Tensor indices = torch::stack(idx);
// counts.index_put_({ indices }, counts.index({ indices }) + 1);
// }
// return counts;
// };
// // Lambda function to compute marginal counts of states
// auto marginalCounts = [](const torch::Tensor& jointCounts) {
// return jointCounts.sum(-1);
// };
@ -113,30 +162,72 @@ namespace bayesnet {
// for (auto& pair : nodes) {
// Node* node = pair.second;
// vector<int> indices;
// for (const auto& parent : node->getParents()) {
// indices.push_back(nodes[parent->getName()]->getId());
// // Create a list of names of the node and its parents
// std::vector<string> nodeNames;
// nodeNames.push_back(node->getName());
// for (Node* parent : node->getParents()) {
// nodeNames.push_back(parent->getName());
// }
// indices.push_back(node->getId());
// for (auto& child : node->getChildren()) {
// torch::Tensor counts = jointCounts(dataset, indices, node->getNumStates()) + laplaceSmoothing;
// // Compute counts and normalize to get probabilities
// torch::Tensor counts = jointCounts(nodeNames) + laplaceSmoothing;
// torch::Tensor parentCounts = marginalCounts(counts);
// parentCounts = parentCounts.unsqueeze(-1);
// torch::Tensor cpt = counts.to(torch::kDouble) / parentCounts.to(torch::kDouble);
// setCPD(node->getCPDKey(child), cpt);
// // The CPT is represented as a tensor and stored in the Node
// node->setCPT((counts.to(torch::kDouble) / parentCounts.to(torch::kDouble)));
// }
// }
void Network::estimateParameters()
{
// Lambda function to compute joint counts of states
auto jointCounts = [this](const vector<string>& nodeNames) {
int size = nodeNames.size();
std::vector<int64_t> sizes(size);
for (int i = 0; i < size; ++i) {
sizes[i] = this->nodes[nodeNames[i]]->getNumStates();
}
torch::Tensor& Network::getCPD(const string& key)
{
return cpds[key];
torch::Tensor counts = torch::zeros(sizes, torch::kLong);
int dataSize = this->dataset[nodeNames[0]].size();
torch::List<c10::optional<torch::Tensor>> indices;
for (int dataIdx = 0; dataIdx < dataSize; ++dataIdx) {
indices.clear();
for (int i = 0; i < size; ++i) {
indices.push_back(torch::tensor(this->dataset[nodeNames[i]][dataIdx], torch::kLong));
}
//torch::Tensor indices = torch::stack(idx);
counts.index_put_({ indices }, counts.index({ indices }) + 1);
}
void Network::setCPD(const string& key, const torch::Tensor& cpt)
{
cpds[key] = cpt;
return counts;
};
// Lambda function to compute marginal counts of states
auto marginalCounts = [](const torch::Tensor& jointCounts) {
return jointCounts.sum(-1);
};
for (auto& pair : nodes) {
Node* node = pair.second;
// Create a list of names of the node and its parents
std::vector<string> nodeNames;
nodeNames.push_back(node->getName());
for (Node* parent : node->getParents()) {
nodeNames.push_back(parent->getName());
}
// Compute counts and normalize to get probabilities
torch::Tensor counts = jointCounts(nodeNames) + laplaceSmoothing;
torch::Tensor parentCounts = marginalCounts(counts);
parentCounts = parentCounts.unsqueeze(-1);
// The CPT is represented as a tensor and stored in the Node
node->setCPT((counts.to(torch::kDouble) / parentCounts.to(torch::kDouble)));
}
}
}

11
Network.h
View File

@ -3,12 +3,16 @@
#include "Node.h"
#include <map>
#include <vector>
namespace bayesnet {
class Network {
private:
map<string, Node*> nodes;
map<string, torch::Tensor> cpds; // Map from CPD key to CPD tensor
map<string, vector<int>> dataset;
Node* root;
vector<string> features;
string className;
int laplaceSmoothing;
bool isCyclic(const std::string&, std::unordered_set<std::string>&, std::unordered_set<std::string>&);
public:
@ -19,9 +23,8 @@ namespace bayesnet {
void addEdge(const string, const string);
map<string, Node*>& getNodes();
void fit(const vector<vector<int>>&, const vector<int>&, const vector<string>&, const string&);
void buildNetwork(const vector<vector<int>>&, const vector<int>&, const vector<string>&, const string&);
torch::Tensor& getCPD(const string&);
void setCPD(const string&, const torch::Tensor&);
void estimateParameters();
void buildNetwork();
void setRoot(string);
Node* getRoot();
};

8
Node.cc
View File

@ -41,8 +41,12 @@ namespace bayesnet {
{
return numStates;
}
string Node::getCPDKey(const Node* child) const
torch::Tensor& Node::getCPT()
{
return name + "-" + child->getName();
return cpt;
}
void Node::setCPT(const torch::Tensor& cpt)
{
this->cpt = cpt;
}
}

2
Node.h
View File

@ -12,6 +12,7 @@ namespace bayesnet {
string name;
vector<Node*> parents;
vector<Node*> children;
torch::Tensor cpTable;
int numStates;
torch::Tensor cpt;
public:
@ -27,7 +28,6 @@ namespace bayesnet {
void setCPT(const torch::Tensor&);
int getNumStates() const;
int getId() const { return id; }
string getCPDKey(const Node*) const;
};
}
#endif

22
main.cc
View File

@ -11,10 +11,15 @@ using namespace std;
vector<mdlp::labels_t> discretize(vector<mdlp::samples_t>& X, mdlp::labels_t& y)
{
vector<mdlp::labels_t>Xd;
auto fimdlp = mdlp::CPPFImdlp();
for (int i = 0; i < X.size(); i++) {
fimdlp.fit(X[i], y);
Xd.push_back(fimdlp.transform(X[i]));
mdlp::labels_t& xd = fimdlp.transform(X[i]);
cout << "X[" << i << "]: ";
auto mm = minmax_element(xd.begin(), xd.end());
cout << *mm.first << " " << *mm.second << endl;
Xd.push_back(xd);
}
return Xd;
}
@ -33,7 +38,7 @@ int main()
features.push_back(feature.first);
}
// Discretize Dataset
vector<mdlp::labels_t> Xd = discretize(X, y);;
vector<mdlp::labels_t> Xd = discretize(X, y);
// Build Network
auto network = bayesnet::Network();
network.fit(Xd, y, features, className);
@ -53,6 +58,19 @@ int main()
cout << "Root: " << network.getRoot()->getName() << endl;
network.setRoot(className);
cout << "Now Root should be class: " << network.getRoot()->getName() << endl;
cout << "CPDs:" << endl;
auto nodes = network.getNodes();
auto classNode = nodes[className];
for (auto it = nodes.begin(); it != nodes.end(); it++) {
cout << "* Name: " << it->first << " " << it->second->getName() << " -> " << it->second->getNumStates() << endl;
cout << "Parents: ";
for (auto parent : it->second->getParents()) {
cout << parent->getName() << " -> " << parent->getNumStates() << ", ";
}
cout << endl;
auto cpd = it->second->getCPT();
cout << cpd << endl;
}
cout << "PyTorch version: " << TORCH_VERSION << endl;
return 0;
}

66
test.cc
View File

@ -1,23 +1,53 @@
#include <map>
#include <string>
#include <iostream>
// #include <torch/torch.h>
using namespace std;
// int main()
// {
// torch::Tensor t = torch::rand({ 5, 5 });
int main(int argc, char const* argv[])
{
map<string, int> m;
m["a"] = 1;
m["b"] = 2;
m["c"] = 3;
if (m.find("b") != m.end()) {
cout << "Found b" << endl;
} else {
cout << "Not found b" << endl;
}
// for (auto [key, value] : m) {
// cout << key << " " << value << endl;
// // Print original tensor
// std::cout << t << std::endl;
// // New value
// torch::Tensor new_val = torch::tensor(10.0f);
// // Indices for the cell you want to update
// auto index_i = torch::tensor({ 2 });
// auto index_j = torch::tensor({ 3 });
// // Update cell
// t.index_put_({ index_i, index_j }, new_val);
// // Print updated tensor
// std::cout << t << std::endl;
// }
#include <torch/torch.h>
return 0;
int main()
{
torch::Tensor t = torch::rand({ 5, 4, 3 }); // 3D tensor for this example
int i = 3, j = 1, k = 2; // Indices for the cell you want to update
// Print original tensor
std::cout << t << std::endl;
// New value
torch::Tensor new_val = torch::tensor(10.0f);
// Indices for the cell you want to update
std::vector<torch::Tensor> indices;
indices.push_back(torch::tensor(i)); // Replace i with your index for the 1st dimension
indices.push_back(torch::tensor(j)); // Replace j with your index for the 2nd dimension
indices.push_back(torch::tensor(k)); // Replace k with your index for the 3rd dimension
//torch::ArrayRef<at::indexing::TensorIndex> indices_ref(indices);
// Update cell
//torch::Tensor result = torch::stack(indices);
//torch::List<c10::optional<torch::Tensor>> indices_list = { torch::tensor(i), torch::tensor(j), torch::tensor(k) };
torch::List<c10::optional<torch::Tensor>> indices_list;
indices_list.push_back(torch::tensor(i));
indices_list.push_back(torch::tensor(j));
indices_list.push_back(torch::tensor(k));
//t.index_put_({ torch::tensor(i), torch::tensor(j), torch::tensor(k) }, new_val);
t.index_put_(indices_list, new_val);
// Print updated tensor
std::cout << t << std::endl;
}