Sunday, October 30, 2011

MAS assignment 2 deadline

There was some confusion over the deadline for the second MAS assignment, about whether it was final or points get deducted if you are late. I sent an email to Matthijs and he just replied: 
0.5 grade point per day
So, to re-iterate, assignment 2 is just like assignment 1; you get a penalty of half a point for each day after the deadline you are late. So, if anyone feels that one more day can really make a difference (like me), then this is good news.

15 comments:

  1. The hard deadline with point deduction really makes me feel like I am a master student at an academic university


    anyhow: I hope I get everything working tonight but it is a pretty big assignment with long training times and difficult debugging.

    ReplyDelete
  2. What do you mean by saying "The hard deadline with point deduction really makes me feel like I am a master student at an academic university"? :S

    All this minutia with points here and points there is beginning to get on my nerves really. Science takes time, it's not a freaking race.

    ReplyDelete
  3. oh lol, I surrounded it with < sarcasm > tags, but forgot to check whether they were actually visible :P

    ReplyDelete
  4. It's HTML. It was written to ignore tags it does not understand. Sadly, the internet does not understand sarcasm :(

    ReplyDelete
  5. Off-topic, but whatever: what capture times are you guys getting? We got 15.76 for the first one, and 28.72 for the second one. Something tells me this is a shitty result ;)

    ReplyDelete
  6. nah, it is to be expected. It depends a bit on your specific settings and whether or not you train the entire state space or not. I have the first one at 18 cycles and am running the second one.

    One hour to go :P

    ReplyDelete
  7. Well the best I got was 21 for the first one and I'm still working on the second one. I have a history of shitty results, but you're probably okay with what you got.

    ReplyDelete
  8. How big is the state space you are training in (or are you training on the whole state space)

    also: after how many episodes did you get your results?

    ReplyDelete
  9. We have about 15.2 on the first one... Second one only 32-33ish... We suck >.<

    ReplyDelete
  10. for us are 11.2 and 14 :P But we needed a huge (~50.000) number of training episodes. Training on a 7x7 state space

    ReplyDelete
  11. As Robrecht implied above, we really shouldn't be focusing on capture times but at the constraints we used and the learning rate we achieved. A capture time in the hundreds if you train with the whole statespace and only applying rewards at the end of the episode is probably better than a capture time in the low twenties if you train in a 3x3 square and give out rewards inverse than the distance to the prey.

    ReplyDelete
  12. Apart from the fact that 20 points of our grades is determined based on our capture times relative to the other students...

    ReplyDelete
  13. We need to get liberated from the grades! Viva la revolution! :p

    ReplyDelete