Implement the functionality of the hyperparameter alpha_block with test

CHANGELOG.md

@@ -7,6 +7,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- Add a new hyperparameter to the BoostAODE class, *alphablock*, to control the way α is computed, with the last model or with the ensmble built so far. Default value is *false*.
+
 ## [1.0.6] 2024-11-23
 
 ### Fixed

bayesnet/ensembles/Boost.cc

@@ -12,7 +12,7 @@
 namespace bayesnet {
     Boost::Boost(bool predict_voting) : Ensemble(predict_voting)
     {
-        validHyperparameters = { "order", "convergence", "convergence_best", "bisection", "threshold", "maxTolerance",
+        validHyperparameters = { "alpha_block", "order", "convergence", "convergence_best", "bisection", "threshold", "maxTolerance",
         "predict_voting", "select_features", "block_update" };
     }
     void Boost::setHyperparameters(const nlohmann::json& hyperparameters_)
@@ -26,6 +26,10 @@ namespace bayesnet {
             }
             hyperparameters.erase("order");
         }
+        if (hyperparameters.contains("alpha_block")) {
+            alpha_block = hyperparameters["alpha_block"];
+            hyperparameters.erase("alpha_block");
+        }
         if (hyperparameters.contains("convergence")) {
             convergence = hyperparameters["convergence"];
             hyperparameters.erase("convergence");
@@ -66,6 +70,12 @@ namespace bayesnet {
             block_update = hyperparameters["block_update"];
             hyperparameters.erase("block_update");
         }
+        if (block_update && alpha_block) {
+            throw std::invalid_argument("alpha_block and block_update cannot be true at the same time");
+        }
+        if (block_update && !bisection) {
+            throw std::invalid_argument("block_update needs bisection to be true");
+        }
         Classifier::setHyperparameters(hyperparameters);
     }
     void Boost::buildModel(const torch::Tensor& weights)

bayesnet/ensembles/Boost.h

@@ -45,8 +45,8 @@ namespace bayesnet {
         std::string select_features_algorithm = Orders.DESC; // Selected feature selection algorithm
         FeatureSelect* featureSelector = nullptr;
         double threshold = -1;
-        bool block_update = false;
+        bool block_update = false; // if true, use block update algorithm, only meaningful if bisection is true
-
+        bool alpha_block = false; // if true, the alpha is computed with the ensemble built so far and the new model
     };
 }
 #endif

bayesnet/ensembles/BoostAODE.cc

@@ -92,7 +92,25 @@ namespace bayesnet {
                 model->fit(dataset, features, className, states, weights_, smoothing);
                 alpha_t = 0.0;
                 if (!block_update) {
-                    auto ypred = model->predict(X_train);
+                    torch::Tensor ypred;
+                    if (alpha_block) {
+                        //
+                        // Compute the prediction with the current ensemble + model
+                        //
+                        // Add the model to the ensemble
+                        n_models++;
+                        models.push_back(std::move(model));
+                        significanceModels.push_back(1);
+                        // Compute the prediction
+                        ypred = predict(X_train);
+                        // Remove the model from the ensemble
+                        model = std::move(models.back());
+                        models.pop_back();
+                        significanceModels.pop_back();
+                        n_models--;
+                    } else {
+                        ypred = model->predict(X_train);
+                    }
                     // Step 3.1: Compute the classifier amout of say
                     std::tie(weights_, alpha_t, finished) = update_weights(y_train, ypred, weights_);
                 }

tests/TestBoostAODE.cc

@@ -130,6 +130,8 @@ TEST_CASE("Oddities", "[BoostAODE]")
         { { "select_features","IWSS" }, { "threshold", 0.51 } },
         { { "select_features","FCBF" }, { "threshold", 1e-8 } },
         { { "select_features","FCBF" }, { "threshold", 1.01 } },
+        { { "alpha_block", true }, { "block_update", true } },
+        { { "bisection", false }, { "block_update", true } },
     };
     for (const auto& hyper : bad_hyper_fit.items()) {
         INFO("BoostAODE hyper: " << hyper.value().dump());
@@ -137,7 +139,6 @@ TEST_CASE("Oddities", "[BoostAODE]")
         REQUIRE_THROWS_AS(clf.fit(raw.Xv, raw.yv, raw.features, raw.className, raw.states, raw.smoothing), std::invalid_argument);
     }
 }
-
 TEST_CASE("Bisection Best", "[BoostAODE]")
 {
     auto clf = bayesnet::BoostAODE();
@@ -180,7 +181,6 @@ TEST_CASE("Bisection Best vs Last", "[BoostAODE]")
     auto score_last = clf.score(raw.X_test, raw.y_test);
     REQUIRE(score_last == Catch::Approx(0.976666689f).epsilon(raw.epsilon));
 }
-
 TEST_CASE("Block Update", "[BoostAODE]")
 {
     auto clf = bayesnet::BoostAODE();
@@ -211,3 +211,18 @@ TEST_CASE("Block Update", "[BoostAODE]")
     // }
     // std::cout << "Score " << score << std::endl;
 }
+TEST_CASE("Alphablock", "[BoostAODE]")
+{
+    auto clf_alpha = bayesnet::BoostAODE();
+    auto clf_no_alpha = bayesnet::BoostAODE();
+    auto raw = RawDatasets("diabetes", true);
+    clf_alpha.setHyperparameters({
+        {"alpha_block", true},
+        });
+    clf_alpha.fit(raw.X_train, raw.y_train, raw.features, raw.className, raw.states, raw.smoothing);
+    clf_no_alpha.fit(raw.X_train, raw.y_train, raw.features, raw.className, raw.states, raw.smoothing);
+    auto score_alpha = clf_alpha.score(raw.X_test, raw.y_test);
+    auto score_no_alpha = clf_no_alpha.score(raw.X_test, raw.y_test);
+    REQUIRE(score_alpha == Catch::Approx(0.720779f).epsilon(raw.epsilon));
+    REQUIRE(score_no_alpha == Catch::Approx(0.733766f).epsilon(raw.epsilon));
+}