[Prev][Next][Index][Thread]
Re: AI and mice
While I agree solving a known maze is deterministic I think a NN approach
could be very valuable in the search phase. I am not aware of anyone
claiming to have an optimal searching strategy and I would be very interested
in any ideas people have. Would it be possible to establish the minimum
number of squares that need to be visited to be sure the fastest route has
been found? This could be a test of intelligent searching.
I have my own non-NN ideas for advanced searching but have not explored them
yet because the scoring system divisor of 30 for search times does not reward
intelligent searching enough. I believe the divisor used to be lower and I
do not know why 30 was selected. Does anyone know?
I have a theory that the adjusted scores for every run of an averagely
intelligent mouse should be constant. That is, the faster later runs
(including the scaled search time score) should get the same score as earlier
runs. This should mean that more intelligent mice achieve reducing scores
and less intelligent mice get increasing scores. I looked at this on some
old data and decided that a divisor of 7 for search times would be more
appropriate than 30. A change from 30 to 7 would certainly give a more
intelligent mouse more chance, although an intelligent fast mouse would
obviously still win.