mardi 23 novembre 2010

Go Spurs Go!

I was in Texas in November to attend INFORMS. Precisely, I was in Austin. There were a lot of people: more than 4,000! The hotel was crowded...
Unfortunately, there were only few people attending the 2 sessions about Constraint Programming. I think about 15 people were in the room for each session. However, I found the talks quite interesting and this opinion was share by some other persons. I particularly appreciated the talk given by John Hooker about MDD based solvers. Nevertheless, I think that John should really try to combine some constraints like he started to do with the among constraints instead of claiming that a whole solver can be based on MDD. The use of MDD for combining some constraints sharing only few variables seems to be a very promising idea.
Personnaly, I gave a talk about nurse scheduling, I changed a little bit my CPAIOR talk and it seems that people appreciated it. I also met Ashish Sabharwal who just moved from Cornell to IBM Watson Research Center (near New York city). It seems that some other people are moving there, which has been confirmed by several other persons at IBM. I will investigate more if I have time.

If you want to find more information you can look at Michael Trick's blog where you will find a picture about the Twitters group at Informs (
On my side, I have also a picture to show. Before, going to Austin I went to San Antonio. I have been really well welcome as you can see on this picture

Everybody can recognize the Spurs because it is written on the Pom-Pom girl dresses...
It was my first NBA game and I really enjoyed it! Tony Parker is a really good player!

jeudi 18 novembre 2010

Vehicle Routing Library

Vincent Furnon, who moved from ILOG to Google, has just released a vehicle routing library.
This means that you can really save money by using some products instead some others. In addition, some nice and clever people will be very happy to help you to develop your application whereas some others will see more and more the life in dark blue.

Here is his presentation about his new development. I really encourage you to test it.

"We just pushed the vehicle routing extension of the CP Solver. It's a 
layer above the CP Solver which lets you model a wide range of vehicle 
routing problems from the Traveling Salesman Problem (and its 
variants, ATSP, TSPTW, ...) to multi-vehicle problems with dimension 
constraints (capacities, time windows) and various "routing" 
constraints (optional nodes, alternate nodes,...). Given it exposes 
the CP variables one can extend the model using the constraints 
available in the CP Solver. 

Concerning the resolution of the problem, most of the "cabling" is
hidden so you just need to call the Solve() method on the RoutingModel
class. However the search can be parametrized using command-line
flags. We are mainly using CP-based local search and large
neighborhood search using routing-specific neighborhoods.
Implementations of Tabu Search and Guided Local Search are available
too and have proven to give good results (especially GLS). 
The basic library is in C++ and a Python-wrapped version is available
Documentation will be added in the future but you can have a look at
the TSP example (both in C++ and Python) as a starting point. We'll be
adding a CVRPTW example very shortly. "