Added missing comment about the QP needing to be convex.

--HG--
extra : convert_revision : svn%3Afdd8eb12-d10e-0410-9acb-85c331704f74/trunk%403912

dlib/optimization/optimization_solve_qp_using_smo.h

@@ -28,6 +28,7 @@ namespace dlib
             subject to the following constraints:
                 - trans(e)*alpha == C (i.e. the sum of alpha values doesn't change)
                 - min(alpha) >= 0 (i.e. all alpha values are nonnegative)
+            Where f is convex.  This means that Q should be positive-semidefinite.
 
 
         To get from this problem formulation to the algorithm below we have to 

dlib/optimization/optimization_solve_qp_using_smo_abstract.h

@@ -40,6 +40,7 @@ namespace dlib
                 subject to the following constraints:
                     - sum(alpha) == C (i.e. the sum of alpha values doesn't change)
                     - min(alpha) >= 0 (i.e. all alpha values are nonnegative)
+                Where f is convex.  This means that Q should be positive-semidefinite.
             - The solution to the above QP will be stored in #alpha.
             - This function uses a simple implementation of the sequential minimal
               optimization algorithm.  It starts the algorithm with the given alpha