# Week 9

The algorithm used for running our linear program on a fine resolution grid, as I mentioned, takes an incredibly long time to run. They often weren’t completing on my local computer, so I resorted to remotely accessing the lab computers, which improved run-times. In this case, though, the improvement was from ‘Incomplete’ to ‘2 to 3 hours,’ which also wasn’t ideal. But I was able to get enough data by Monday, after running these programs in the background over the weekend, to draw up statistics and tables to compare the heuristic’s performance against the grid performance. Basically, what we found was that the grid solutions gave *slightly* better dwell-times, but took 3 or 4 orders of magnitude the time to run. This is good news for the heuristic! It seems reasonable to use it to get good solutions quickly.

Next, I formulated the linear program to maximize the number of surfaces completely irradiated with some amount of bounded time allowed. It’s been a while since I’ve done just theory work, as opposed to implementing something. I do enjoy implementation, and being able to see something come out of your work, but my heart is with theoretical work much of the time, so I enjoyed being able to do this for a bit. This formulation turned out to be a mixed integer linear program as well.

This week was the RSS conference.

There’s not much to say for this week besides this, though. After this formulation, and besides a number of meetings (which seem to keep increasing as the weeks go by!) I spent most of time focused on implementing the above MILP, as well as the one that minimizes total disinfection time that I talked about last week. To implement these, I’ve been using a great optimization solver called Gurobi and it works a lot better than the linear programming software I was using before, so I think I’d like to port over the old code to use Gurobi instead.