Adding noise type kernel

include/kernel_functors.hpp

@@ -55,6 +55,9 @@ namespace bayesopt
     virtual double gradient( const vectord &x1, const vectord &x2,
 			     size_t component ) = 0;
 
+    // Only kernels with noise modify them.
+    virtual double getNoise(){ return 0.0; };
+
   protected:
     size_t n_inputs;
   };

include/kernels/kernel_const.hpp

@@ -50,6 +50,24 @@ namespace bayesopt
     { return 0.0; };
   };
 
+  class NoiseKernel: public AtomicKernel
+  {
+  public:
+    void init(size_t input_dim)
+    { n_params = 1; n_inputs = input_dim;   };
+
+    double operator()(const vectord &x1, const vectord &x2)
+    { return 0.0; };
+
+    double gradient(const vectord &x1, const vectord &x2,
+		    size_t component)
+    { return 0.0; };
+
+    double getNoise()
+    { return params(0); }
+  };
+
+
   //@}
 
 } //namespace bayesopt

src/kernel_functors.cpp

@@ -45,6 +45,7 @@ namespace bayesopt
   KernelFactory::KernelFactory()
   {
     registry["kConst"] = & create_func<ConstKernel>;
+    registry["kNoise"] = & create_func<NoiseKernel>;
     registry["kLinear"] = & create_func<LinKernel>;
     registry["kLinearARD"] = & create_func<LinKernelARD>;
 
@@ -169,7 +170,7 @@ namespace bayesopt
 	    corrMatrix(ii,jj) = (*mKernel)(XX[ii], XX[jj]);
 	    corrMatrix(jj,ii) = corrMatrix(ii,jj);
 	  }
-	corrMatrix(ii,ii) = (*mKernel)(XX[ii],XX[ii]) + nugget;
+        corrMatrix(ii,ii) = (*mKernel)(XX[ii],XX[ii]) + nugget + mKernel->getNoise();
       }
   }
 

src/kernelregressor.cpp

@@ -53,7 +53,7 @@ namespace bayesopt
   {
     const vectord lastX = mData.getLastSampleX();
     vectord newK = computeCrossCorrelation(lastX);
-    newK(newK.size()-1) += mRegularizer;   // We add it to the last element
+    newK(newK.size()-1) += mRegularizer + mKernel.getKernel()->getNoise();   // We add it to the last element
     utils::cholesky_add_row(mL,newK);
     precomputePrediction(); 
   } // updateSurrogateModel