change for loop into numpy functions

[paul18fr]

It's look like results coming from FEM

My comments:

• "append" is your enemy (dynamic allocation memory = time consuming)
  
• "numpy.unique" provides you a new matrix without double node values ⇒ then "numpy.size" gives you the number of rows of your new matrix ⇒ use "numpy.zeros" to reserve in memory the matrix (gain of time)
  
• sorting nodes numbers is a good idea: the second node follows immediatly the first one ⇒ you would be able to sum "i" index with "i+1" one (so I think vectorization use is possible)
  

I do not use Pandas, but frames seem to be no more no less that disctionnaries (Keys + values), so it is easy to extract it and to work with, isn't it?