G3: 13 queries for all 3 random mapping. Could be improved by choosing different group size.
Comments: Optimal strategy should be dynamic, according to the information learnt from query feedbacks. Following a fixed strategy is not likely an optimal one. Or switch between several fixed strategies?
G5: How to calculate the size of the search space? Whether strategy that minimize search space size greedily result in missing the glocal optimal strategy? Try to find orthoganal questions and query them together. How to update former rules when new information is learnt (something like intersection with former rules)?
G2: Boolean vector for each index, then take orthoganal vectors (vectors sharing the least possibility overlap)
G7: Quickly go through all numbers (random decide how to split in half). Then repeat this recursively while shrinking the query size. What is the best shrinking rate? How many to overlap with former queries?
(Didn't follow the discussions over all the groups since helping G2 set up. Sorry...)
For pairwise query strategy, try this: [1,2,2,1,1,2,...]
Things to work on:
How to calculate the search space. How to shrink it most.
Best strategy so far: "BE LUCKY" :)
Next deliverable:
Implement a mapper. Improve the guesser.
No comments:
Post a Comment