Merge branch 'normalize_descent' into 'master'

normalize descent direction for SteepestDescent and VL_BFGS (but not for RelaxedNewton!)

See merge request !188

nifty/minimization/line_searching/line_search_strong_wolfe.py

@@ -121,7 +121,7 @@ class LineSearchStrongWolfe(LineSearch):
             if alpha1 < 0:
                 alpha1 = 1.0
         else:
-            alpha1 = 1.0
+            alpha1 = 1.0/pk.norm()
 
         # start the minimization loop
         for i in xrange(self.max_iterations):