apply code formatting and add some explanation

RecoTauTag/RecoTau/plugins/DeepTauId.cc

@@ -907,9 +907,13 @@ class DeepTauId : public deep_tau::DeepTauBase {
         get(dnn::neutralIsoPtSumdR03_over_dR05) = getValue(tau.tauID("neutralIsoPtSumdR03") / tau.tauID("neutralIsoPtSum"));
         get(dnn::photonPtSumOutsideSignalCone) = getValueNorm(tau.tauID("photonPtSumOutsideSignalConedR03"), 1.731f, 6.846f);
         get(dnn::puCorrPtSum) = getValueNorm(tau.tauID("puCorrPtSum"), 22.38f, 16.34f);
-        get(dnn::tau_dxy_pca_x) = 0; // getValueNorm(tau.dxy_PCA().x(), -0.0241f, 0.0074f);
-        get(dnn::tau_dxy_pca_y) = 0; // getValueNorm(tau.dxy_PCA().y(),0.0675f, 0.0128f);
-        get(dnn::tau_dxy_pca_z) = 0; // getValueNorm(tau.dxy_PCA().z(), 0.7973f, 3.456f);
+        // The global PCA coordinates were used as inputs during the NN training, but it was decided to disable
+        // them for the inference, because modeling of dxy_PCA in MC poorly describes the data, and x and y coordinates
+        // in data results outside of the expected 5 std. dev. input validity range. On the other hand,
+        // these coordinates are strongly era-dependent. Kept as comment to document what NN expects.
+        get(dnn::tau_dxy_pca_x) = 0;  // getValueNorm(tau.dxy_PCA().x(), -0.0241f, 0.0074f);
+        get(dnn::tau_dxy_pca_y) = 0;  // getValueNorm(tau.dxy_PCA().y(),0.0675f, 0.0128f);
+        get(dnn::tau_dxy_pca_z) = 0;  // getValueNorm(tau.dxy_PCA().z(), 0.7973f, 3.456f);
 
         const bool tau_dxy_valid = std::isnormal(tau.dxy()) && tau.dxy() > - 10 && std::isnormal(tau.dxy_error())
             && tau.dxy_error() > 0;