Add Convergence hyperparameter

.vscode/launch.json

@@ -25,9 +25,9 @@
             "program": "${workspaceFolder}/build/src/Platform/main",
             "args": [
                 "-m",
-                "AODE",
+                "BoostAODE",
                 "-p",
-                "/home/rmontanana/Code/discretizbench/datasets",
+                "/Users/rmontanana/Code/discretizbench/datasets",
                 "--stratified",
                 "-d",
                 "mfeat-morphological",
@@ -35,7 +35,7 @@
                 // "--hyperparameters",
                 // "{\"repeatSparent\": true, \"maxModels\": 12}"
             ],
-            "cwd": "/home/rmontanana/Code/discretizbench",
+            "cwd": "/Users/rmontanana/Code/discretizbench",
         },
         {
             "type": "lldb",

src/BayesNet/BoostAODE.cc

@@ -3,6 +3,7 @@
 #include "BayesMetrics.h"
 #include "Colors.h"
 #include "Folding.h"
+#include <limits.h>
 
 namespace bayesnet {
     BoostAODE::BoostAODE() : Ensemble() {}
@@ -13,7 +14,7 @@ namespace bayesnet {
     void BoostAODE::setHyperparameters(nlohmann::json& hyperparameters)
     {
         // Check if hyperparameters are valid
-        const vector<string> validKeys = { "repeatSparent", "maxModels", "ascending" };
+        const vector<string> validKeys = { "repeatSparent", "maxModels", "ascending", "convergence" };
         checkHyperparameters(validKeys, hyperparameters);
         if (hyperparameters.contains("repeatSparent")) {
             repeatSparent = hyperparameters["repeatSparent"];
@@ -24,27 +25,30 @@ namespace bayesnet {
         if (hyperparameters.contains("ascending")) {
             ascending = hyperparameters["ascending"];
         }
+        if (hyperparameters.contains("convergence")) {
+            convergence = hyperparameters["convergence"];
+        }
     }
     void BoostAODE::validationInit()
     {
         auto y_ = dataset.index({ -1, "..." });
         auto fold = platform::StratifiedKFold(5, y_, 271);
-        // save input dataset
         dataset_ = torch::clone(dataset);
+        // save input dataset
         auto [train, test] = fold.getFold(0);
         auto train_t = torch::tensor(train);
         auto test_t = torch::tensor(test);
         // Get train and validation sets
-        X_train = dataset.index({ "...", train_t });
+        X_train = dataset.index({ torch::indexing::Slice(0, dataset.size(0) - 1), train_t });
         y_train = dataset.index({ -1, train_t });
-        X_test = dataset.index({ "...", test_t });
+        X_test = dataset.index({ torch::indexing::Slice(0, dataset.size(0) - 1), test_t });
         y_test = dataset.index({ -1, test_t });
-        // Build dataset with train data
         dataset = X_train;
-        buildDataset(y_train);
         m = X_train.size(1);
         auto n_classes = states.at(className).size();
         metrics = Metrics(dataset, features, className, n_classes);
+        // Build dataset with train data
+        buildDataset(y_train);
     }
     void BoostAODE::trainModel(const torch::Tensor& weights)
     {
@@ -56,9 +60,18 @@ namespace bayesnet {
         Tensor weights_ = torch::full({ m }, 1.0 / m, torch::kFloat64);
         bool exitCondition = false;
         unordered_set<int> featuresUsed;
+        // Variables to control the accuracy finish condition
+        double priorAccuracy = 0.0;
+        double delta = 1.0;
+        double threshold = 1e-4;
+        int tolerance = convergence ? 5 : INT_MAX; // number of times the accuracy can be lower than the threshold
+        int count = 0; // number of times the accuracy is lower than the threshold
+        fitted = true; // to enable predict
         // Step 0: Set the finish condition
         // if not repeatSparent a finish condition is run out of features
         // n_models == maxModels
+        // epsiolon sub t > 0.5 => inverse the weights policy
+        // validation error is not decreasing
         while (!exitCondition) {
             // Step 1: Build ranking with mutual information
             auto featureSelection = metrics.SelectKBestWeighted(weights_, ascending, n); // Get all the features sorted
@@ -81,7 +94,6 @@ namespace bayesnet {
             }
             featuresUsed.insert(feature);
             model = std::make_unique<SPODE>(feature);
-            n_models++;
             model->fit(dataset, features, className, states, weights_);
             auto ypred = model->predict(X_train);
             // Step 3.1: Compute the classifier amout of say
@@ -102,12 +114,22 @@ namespace bayesnet {
             // Step 3.4: Store classifier and its accuracy to weigh its future vote
             models.push_back(std::move(model));
             significanceModels.push_back(alpha_t);
-            exitCondition = n_models == maxModels && repeatSparent || epsilon_t > 0.5;
+            n_models++;
+            auto y_val_predict = predict(X_test);
+            double accuracy = (y_val_predict == y_test).sum().item<double>() / (double)y_test.size(0);
+            if (priorAccuracy == 0) {
+                priorAccuracy = accuracy;
+            } else {
+                delta = accuracy - priorAccuracy;
+            }
+            if (delta < threshold) {
+                count++;
+            }
+            exitCondition = n_models == maxModels && repeatSparent || epsilon_t > 0.5 || count > tolerance;
         }
         if (featuresUsed.size() != features.size()) {
             status = WARNING;
         }
-        weights.copy_(weights_);
     }
     vector<string> BoostAODE::graph(const string& title) const
     {

src/BayesNet/BoostAODE.h

@@ -19,6 +19,7 @@ namespace bayesnet {
         bool repeatSparent = false;
         int maxModels = 0;
         bool ascending = false; //Process KBest features ascending or descending order
+        bool convergence = false; //if true, stop when the model does not improve
     };
 }
 #endif

src/BayesNet/Ensemble.cc

@@ -24,7 +24,7 @@ namespace bayesnet {
             // i.e. votes[0] contains how much value has the value 0 of class. That value is generated by the models predictions
             vector<double> votes(numClasses, 0.0);
             for (int j = 0; j < n_models; ++j) {
-                votes[y_pred_[i][j]] += significanceModels[j];
+                votes[y_pred_[i][j]] += significanceModels.at(j);
             }
             // argsort in descending order
             auto indices = argsort(votes);