Very interesting, but the code example used in the documentation is pretty poor:

    from simpleai.search import SearchProblem, astar

    GOAL = 'HELLO WORLD'

    class HelloProblem(SearchProblem):
        def actions(self, state):
            if len(state) < len(GOAL):
                return list(' ABCDEFGHIJKLMNOPQRSTUVWXYZ')
            else:
                return []

        def result(self, state, action):
            return state + action

        def is_goal(self, state):
            return state == GOAL

        def heuristic(self, state):
            # how far are we from the goal?
            wrong = sum([1 if state[i] != GOAL[i] else 0
                        for i in range(len(state))])
            missing = len(GOAL) - len(state)
            return wrong + missing

Among other things, why create a class and hardcode the GOAL as a constant?

It's just illustrative, to understand how you give the information needed to run one of the algorithms.

May be we can think of a better example, will try :)

besides, not always you want to find the goal state. Many problems are about findig the path of actions from the initial state to the goal state, both known.

This is awesome. I really want to play with some AI lib and all of the ones I tried have way too many problems compiling on Windows. pip install them fails too. I don't want to spend time compiling libraries.

This one just worked. This is the Python way.

Have you tried active python? Pypm install has many libraries compiled for windows.

I totally forgot about it. You are right, I might give some of the more hardcore ones after I'm done playing with this one. Thanks!

See also: https://code.google.com/p/aima-java/