Added option to choose power of AE objective

Co-authored-by: eelregit [email protected]

docs/src/man/lhcoptim.md

@@ -11,11 +11,8 @@ algorithm. The objective function is the inverse of the Audze-Eglais function de
     This package maximises the inverse of the Audze-Eglais objective function.
 
 
-where ``L^2_{pq}`` is the Euclidean norm. It is also possible to choose power other than 2
-using the `ae_power` optional parameter.
-!!! note
-
-    Large integer powers can cause overflow leading to the Audze-Eglais objective function to fail silently. Use floats for large powers at the cost of computational speed.
+where ``L_{pq}`` is the Euclidean norm. It is also possible to choose powers other than 2
+by using the `ae_power` optional parameter.
 
 
 There is also an option to use a periodic version of the Audze-Eglais which reduces clustering 

src/AudzeEglaisObjective.jl

@@ -2,7 +2,7 @@
     n,d = size(LHC)
     dist = 0.0
 
-    # Squared l-2 norm of distances between all (unique) points
+    #l-2 norm of distances between all (unique) points to the power of ae_power
     for i = 2:n
         for j = 1:i-1
             dist_tmp = 0.0
@@ -13,9 +13,9 @@
                 else
                     @inbounds dist_comp = LHC[i,k]-LHC[j,k]
                 end
-                dist_tmp += dist_comp^ae_power
+                dist_tmp += dist_comp^2
             end
-            dist += 1/dist_tmp
+            dist += 1/dist_tmp^(ae_power/2)
         end
     end
     output = 1/dist