fixup! [ML] Randomize train/test cluster boundary assignment in RDataLoader

Author: siliataider

tree/ml/inc/ROOT/ML/RClusterLoader.hxx

@@ -401,11 +401,14 @@ public:
             std::min(static_cast<std::size_t>(totalFiltered * (1.0f - fValidationSplit)), trainRemaining);
          const std::size_t valCount = totalFiltered - trainCount;
 
-         // We generate a random boolean value to decide whether the training set gets the prefix
-         // or suffix of each cluster to ensure better shuffling across runs when splitting.
-         std::mt19937 g(fSetSeed + fAccumulatedFilteredForTrain); // vary per cluster
-         std::uniform_int_distribution<int> coin(0, 1);
-         const uint64_t trainIsPrefix = coin(g);
+         uint64_t trainIsPrefix = true;
+         if (fShuffle) {
+            // If shuffling is enabled, we generate a random boolean value to decide whether the training set
+            // gets the prefix or suffix of each cluster to ensure better shuffling across runs when splitting.
+            std::mt19937 g(fSetSeed + fAccumulatedFilteredForTrain); // vary per cluster
+            std::uniform_int_distribution<int> coin(0, 1);
+            trainIsPrefix = coin(g);
+         }
 
          // The boundary is the raw entry index of the first entry assigned to validation.
          // Stable across epochs since the same filter always produces the same ordered entries.