Lecture 5 - Sep. 26

Talk about iSnork.

1. Summarize whole sight area instead of single high value creature. Calculate average happiness and quantize the value into 26 levels.

2. Send message every turn is not necessary given divers are not moving every min. Instead, send messages every 3 mins containing 3 letters (1 per min).

3. Average or most interesting? Especially when you don't have time to see everything.

4. Preference on static creatures in dense map.

5. Use marginal happiness instead of absolute happiness.

6. Spend bits on specifying the position within the sight.

7. Strategy for reaction to messages. Whether it worthes to follow. Assumption of a uniform distribution of species of creatures could be very helpful. But how confident it can be?

8. Map depends on listener

Tracking valuable creatures for others' reference.

Tourament configuration:

Boards: 8 boards

5 seen: Pirhanna, Hidden Treasure, Starwars, Diversity Day, Very Happy

3 unseen: One large creature; one high value, verying many low value dangerous creatures; very many species; order of many scales of happiness

D: 23, 12, 30, 100

R: 6, 3, 15, D * 3

N: 7, 19, 2, 28, 100

Average on ranks.

Disqualification: bottom of rank

Lecture 4 - Sep. 21

Anouncement:

Proj 1 final deadline after one week. Presentation 8 mins for each group.

Starwar map used. Size 15, R, N unchanged. Max happiness: 36,750. Random seed: 1,316,624,910,655

Question: statice creature on board.

G2: 17350. Lots of divers come back to boat early. Danger avoidance stategy can be improved to avoid penalty, especially in early game.

G9: -101600. take 1 hour to explore. ingore and step over dangers in early game based on map danger evaluation. then divers cluster and stay close to corner and edge. Confidence transmitted in messaging. All divers back.

C&Q: How is message from others change the behavior? (message value decay along time).

G8: 34450. spend 7 mins for planning and avoid dangers at beginning. using isnorkle. After seen everying, divers back to boat.

C&Q:

G7: 17,550. Same as G8 in early game. Try to keep a spiral track and avoid danger at the same time result in some clustered divers. Go around a danger then try to go back to the closest point on the spiral. All divers back to boat. But losing points on the way back.

C&Q: what if the danger is too dense for the divers to keep track on spiral. will it result in stuck at local minimum? (suggestion from other group: set time out limit. if wait too long, just random walk.)

G6: -85,450. Jump into the sea right away. But not avoiding danger enough. Hurt and followed by some really dangerous creature. All back to boats at 5h, but returned to the sea again. All back finally.

C&Q: cases that back to boat and come out again may benefit.

G5: -136,750. no avoiding danger due to heapmap bug. all divers back to boat though.

G4: have bug :(

G3: -120,550. Try to merge two strategy. Points not explored yet or place others suggest to go. Not appearing to avoid danger. Divres do not go back to boat.

Deliverables 4: next Monday. Last deliverables.

Lecture 3 - Sep. 19

G2 demo: all divers step south by one step and stay given that they have seen all the species. OutOfMemory error in later stage. Diagonal moving strategy.

C&Q: why all divers head to the same direction? more randomness needed? whether diagonal moving is a good way or actually a limitation, especially given the truth that it takes longer to move diagonally.

G3 demo: good inner and outer circle. half of the divers not back to the boat. the radius is determinded by the dimension.

C&Q: more circles.

G4: avoidance of the mines. some divers see the treasure. but iSnorkle is not implemeted yet to let others know. show a little confusion when exploring for the treature.

C&Q: strategy effect is map dependent. randomness gives good covering. but maybe only suitable for relatively small map. better random scattering by keep record of visited place. can it work with moving dangerous creatures? (Yes if estimating the probability of creature moving directions).

G6: Preset path. then switch path to avoid the danger when encounter. Danger avoidance results in group splitting. divers might be tracked by the dangers. 4 of the divers not back to boat.

C&Q: more efficient way than back tracking when switching? Walk along the edge is not efficient for information collection by seeing only one side. Is not coming back really a choice? (Ken's bias is to enforce coming back to the boat. You voted for disqualification)

G7: good two spirals and divers switching. Has some danger avoidance strategy, but don't have time to show in class.

G8: Stay in boat for 2mins for planning. Good work on danger avoidance. not using iSnorkle.

C&Q: Strategies could be strengthened benefitting from iSnorkling. What if the dangers are really dense? how to balance the need of danger avoidance and the enforcement of getting back to the boat in late game?

G9: Try to use spiral strategies. But haven't been able to give each diver a different spirals.

C&Q: Grouping up divers and send group message with more bits per message.

(It seems that I confused the comments between G4 and G5 by mixing them together or missing one of them. Sorry guys :| )

Lecture 2 - Sep. 14

Discussion and comments on the submitted maps, Whether it is necessary to give special algoritm for extreme cases which can be evaluated by overall map information? :

All except one creature are dangerous. Do the divers really want to jump into the sea or just stay on the boat staying away from all the dangers?

Tons of species

Some really dangerous creatures

Empty map, nothing interesting but engery cost

Offline learning

What to do before Monday (suggestions):

Danger avoidance

Path planning

Map overal analysis and evaluation

Infrastructure that allows new strategy to be plug in (time consuming)

Deliverables for Monday:

Working player, shows some strategies on some preferred the boards. Due by 12:00pm next Monday. Additional maps are welcome. No presentation or report required.

Project 1 Team Assignment

G2: Riddhi, Afreen, Tao, Siyang

G3: William, Hans, Moses, Andrew

G4: Kannapiran, Erik, Alex G.

G5: Alex B. , Colleen, Ben, Siddharth

G6: Jonecia, Elliot, Erica

G7: Dhaval, Ashish, Jonathan

G8: Orestis, Nathaniel, Najaf

G9: Sandeep, Neil, Chris, Yufei

Lecture 1 - Sep. 12

Brought up ideas from last year.

1. 1 hr simply walk around then try to optimize

2. spiral search strategy

3. dangerous creature avoidance

4. get back to the boat in time!

5. evaluate the worth of the message depending both the creature and distance (more details need to be discussed about the distance)

6. assign diver to follow really valuable creature to update with other coming divers. then they choose to switch following and browsing roles.

7. Divers' happiness are not shared. But it can be estimated basing on each one's message history

8. dynamic parameters of Divers

Suggestions:

1. Divide divers into explorers and browsers

2. Chain the divers and let them follow others

3. Sample the map and guess the whole picture?

4. Take advantage of the unlikihood of the creature makeing a turn

5. Explicit message for "seeing nothing" or "no danger"

Clarifications:

1. Euclidean distance, from square center to center

2. Boat can be used as shelter from dangerous creature

3. No credit for observation when in boat

4. Overlap on the same square is allowed (divers, creatures)

Anouncement:

1. Deliverables this Wednesday, dumb players that can correctly call the API as well as some rough description of potential strategies. Think about a new map. One submission each team.