diff --git a/src/BayesNet/CMakeLists.txt b/src/BayesNet/CMakeLists.txt
index f502701..ea530a9 100644
--- a/src/BayesNet/CMakeLists.txt
+++ b/src/BayesNet/CMakeLists.txt
@@ -1,3 +1,4 @@
 include_directories(${BayesNet_SOURCE_DIR}/lib/mdlp)
+include_directories(${BayesNet_SOURCE_DIR}/lib/Files)
 add_library(BayesNet bayesnetUtils.cc Network.cc Node.cc BayesMetrics.cc Classifier.cc KDB.cc TAN.cc SPODE.cc Ensemble.cc AODE.cc TANNew.cc Mst.cc)
-target_link_libraries(BayesNet mdlp "${TORCH_LIBRARIES}")
\ No newline at end of file
+target_link_libraries(BayesNet mdlp arff "${TORCH_LIBRARIES}")
\ No newline at end of file
diff --git a/src/BayesNet/TANNew.cc b/src/BayesNet/TANNew.cc
index f491356..1ef85e3 100644
--- a/src/BayesNet/TANNew.cc
+++ b/src/BayesNet/TANNew.cc
@@ -1,4 +1,5 @@
 #include "TANNew.h"
+#include "ArffFiles.h"
 
 namespace bayesnet {
     using namespace std;
@@ -12,11 +13,13 @@ namespace bayesnet {
         this->features = features;
         this->className = className;
         Xv = vector<vector<int>>();
+        auto Xvf = vector<vector<float>>();
         yv = vector<int>(y.data_ptr<int>(), y.data_ptr<int>() + y.size(0));
         for (int i = 0; i < features.size(); ++i) {
             auto* discretizer = new mdlp::CPPFImdlp();
             auto Xt_ptr = X.index({ i }).data_ptr<float>();
             auto Xt = vector<float>(Xt_ptr, Xt_ptr + X.size(1));
+            Xvf.push_back(Xt);
             discretizer->fit(Xt, yv);
             Xv.push_back(discretizer->transform(Xt));
             auto xStates = vector<int>(discretizer->getCutPoints().size() + 1);
@@ -28,9 +31,48 @@ namespace bayesnet {
         auto yStates = vector<int>(n_classes);
         iota(yStates.begin(), yStates.end(), 0);
         this->states[className] = yStates;
-        /*
-        Hay que discretizar los datos de entrada y luego en predict discretizar también con el mmismo modelo, hacer un transform solamente.
-        */
+        // Now we have standard TAN and now we implement the proposal
+        // order of local discretization is important. no good 0, 1, 2...
+        auto order = model.topological_sort();
+        auto nodes = model.getNodes();
+        bool upgrade = false; // Flag to check if we need to upgrade the model
+        for (auto feature : order) {
+            auto nodeParents = nodes[feature]->getParents();
+            int index = find(features.begin(), features.end(), feature) - features.begin();
+            vector<string> parents;
+            if (parents.size() == 1) continue; // Only has class as parent
+            upgrade = true;
+            transform(nodeParents.begin(), nodeParents.end(), back_inserter(parents), [](const auto& p) {return p->getName(); });
+            // Remove class as parent as it will be added later
+            parents.erase(remove(parents.begin(), parents.end(), className), parents.end());
+            vector<int> indices;
+            transform(parents.begin(), parents.end(), back_inserter(indices), [&](const auto& p) {return find(features.begin(), features.end(), p) - features.begin(); });
+            // Now we fit the discretizer of the feature conditioned on its parents and the class i.e. discretizer.fit(X[index], X[indices] + y)
+            vector<string> yJoinParents;
+            transform(yv.begin(), yv.end(), back_inserter(yJoinParents), [&](const auto& p) {return to_string(p); });
+            for (auto idx : indices) {
+                for (int i = 0; i < Xvf[idx].size(); ++i) {
+                    yJoinParents[i] += to_string(Xv[idx][i]);
+                }
+            }
+            auto arff = ArffFiles();
+            auto yxv = arff.factorize(yJoinParents);
+            discretizers[feature]->fit(Xvf[index], yxv);
+        }
+        if (upgrade) {
+            // Discretize again X with the new fitted discretizers
+            Xv = vector<vector<int>>();
+            for (int i = 0; i < features.size(); ++i) {
+                auto Xt_ptr = X.index({ i }).data_ptr<float>();
+                auto Xt = vector<float>(Xt_ptr, Xt_ptr + X.size(1));
+                Xv.push_back(discretizers[features[i]]->transform(Xt));
+                auto xStates = vector<int>(discretizers[features[i]]->getCutPoints().size() + 1);
+                iota(xStates.begin(), xStates.end(), 0);
+                this->states[features[i]] = xStates;
+            }
+        }
+
+
         TAN::fit(Xv, yv, features, className, this->states);
         return *this;
     }