where \({{\varepsilon _i}}\) is the normal distribution, then it compares \({\eta = 0.01,0.03,0.05}\)\({x_i}\)with η=0.01,0.03,0.0 respectively. Here I plot the there figures in the first raw of Fig3. For the data sets, they used the same number of neighbors, which is k=10. With the increasing noise level , the computed \({{\tau _i}}\)’s get expressed at points with relatively large noise. The quality of the computed ’s can be improved if it increases the number of neighbors as shown on the column (d) in Fig3