Comments (4)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 GuidoDiepen commented Permalink

Your eternal admiration.... Had to find the cite :) Found your name on slide 39 :)

2 JeanFrancoisPuget commented Permalink

Hi Guido,

well done, you're on my hall of fame! :-)
I'll pay you a drink next time I visit Paragon!

3 EdKlotz commented Permalink

> Then there are various ways to make sure the model is always feasible. One way is to move
> wishes to the objective function.
> Another, related approach, is to minimize the number of wishes that aren't met. In this case we
> add a binary variable to each wish expressed as a constraint.

And the second approach always involves solving a MIP, whereas the first may only require solving a continuous model (i.e. if the original model was also continuous). Thus, the second approach may involve significantly more computational effort than the first, including the possibility of an unfinished run in the allotted time. A business person unfamiliar with optimization probably won't make this distinction. Thus, even when you gave a good tool such as ODME that makes optimization accessible to a wide range of users, challenges remain to avoid giving the business users answer they cannot user.

4 JeanFrancoisPuget commented Permalink


you're right.
I was wondering how long I could go on in this blog without making a difference between "easy" problems (eg LP) and "hard" ones (eg MIP). I don't want to explain to business users what NP hard means...
Yet, as you point out, algorithm complexity kicks in easily.