Wednesday, October 13, 2010

10/13: Class Notes

Deliverable today was an automated player. How did we do?

We ran on "Cage.xml" map:
Daniel: Group 1 put all of the lights clustered together in one spot. They found the biggest area and stuck all of their lights there.
G1: 6763
EPK: Group 2 also looked for a large open area to place the collector in, then use a spirograph strategy to place the rest of the lights. They will be chained inwards, but will not necessarily note locations of walls.
G2: 3473
Neetha: We had a lot of bugs, not 100% sure what happened.
G3: 10278
Flavio: We find the largest area, then move diagonally and then horizontally
G4: 1570
David (G5): We look at 10x10 grid units, so we don't necessarily see the obstacles. A local search isn't necessary, need something more global
G5: Doesn't seem like it would ever finish.
Pilunchana (G6): We had special cases, couldn't get one to work here, so we went randomly, but did very poorly.
G6: No termination
DFed (G7): Collector in most open area. Then place lights to have minimal shadow
G7: 799

Why are we limiting how many lights we use? (ex, only using 7 when we have 10)
John: We should add more branches, so that we don't increase timing difficulties but have more lights
Dan: Have extra auxilary chains that aren't directly working with the main chain, but are working as an extra funnel
Neerja: It's really hard to make the frequency work right
Eva: We can take advantage of the walls to break up our sets of lights so that we don't have interacting flows.

We did a central collector manually and collected almost 50% in 261 steps.
Is this strategy optimal?
Dfed: We haven't proven that there isn't a better way to do it on the side
Neerja: There is no particular advantage in where you put the collector (center or side), because it matters mostly where the obstacles are
Eva: Create a partitioned space, closing the gaps between walls with the shortest line. Then find the largest blob, then see how large the rest of the blobs are, then use that to inform your strategy
Daniel: Calculate the minimum distance that each mosquito would have to travel to get to the collector, then use that information
Nitin: If we could look at this as a graph, we could re-use a lot of network flow algorithms that already exist
Blake: You don't need a complete graph representation, just within the boundaries that you need

What's a good fallback strategy?
John: Just put one long chain
Archana: Putting lights in the corner is a good one
Neetha: Putting the lights near the corner is not good because you lose a lot of the light
Archana: We'd use the corners to build a chain inwards
Neerja: You should NEVER pick a position statically

Monday is our final discussion class. Code will be due Weds and we'll start project 3 then too, and have reports and presentations on the 25th

For Monday: Have a very, very robust player.

No comments:

Post a Comment