Marks distribution in midterm

[Subbarao Kambhampati <rao@asu.edu>]

Here are the marks that people got in midterm (sorted in descending order).

I will return the answerbooks in the class today.



out of 95

78; 77; 72; 71; 71; 70; 53; 52; 46

Rao

Common mistakes include:

--Not realizing that conditional effects are not disjunctive p=>q is a 
short cut for if p is ture in the previous state
  q is true in the next state. Non-det effects will have to be about both 
literals in the next state

--not realizing the reason why PG is easiwer to construct (for each level, 
you get at most one instance of any literal or action--instead of
   possibly exponential copies in the progression tree. The time cost 
comes because you are applying each action many times

--not recognizing that if you don't say that at least one action should 
occur per level, there is no real constraint on the values
of the propositions in successive levels..

--not realizing that classical frame axioms can be made to support 
parallelism if we write them for each "subset" of non-interacting actions
rather than for each action... this increases the number of axioms 
exponentially.

--A couple of people made the costly mistake of thinking that causal 
encoding is the same as graphplan encoding. SOme who wrote the causal encoding
missed writing the constraints corresponding to links (the precedence and 
preservation constraints). Almost no one wrote the constraints for
transitivity of the ordering relation (so I didn't bother penalizing for 
that mistake).

--Many people seem to have confused plan length with number of parallel 
steps in the plan (thus getting heuristic values all wrong)
   No one gave much expalnation as to hwo they got the heuristic values 
making partial credit assignment hard...