Some minor errors need to be fixed

[tianyunzhe]

Very good repo! But there are still some errors need to be fixed.

1. PyPOMDP/pypomdp/parsers/env_parser.py

   Line 187 in 3f115b4

   self.Z[(action, next_state, obs)] = prob

   
   to self.Z[(action, next_state, obs)] = float(prob)

2. PyPOMDP/pypomdp/models/model.py

   Lines 32 to 34 in 3f115b4

   	@property
   	def num_actions(self):
   	return len(self.states)

   
   to return len(self.actions)

[tianyunzhe]

Besieds, in 'pbvi', it seems like that you don't use the belief point set expansion proposed by the original paper. Here is my code to expand the belief point set followed the description by that paper, hope it can help you.

 def belief_point_expansion(self):
        for belief_state in self.belief_points:
            init_state = self.model.states[draw_args(belief_state)]
            largest_distance = 0
            add_belief = None
            for action in self.model.actions:
                sj, oj, reward, cost = self.model.simulate_action(init_state, action)
                new_belief = self.update_belief(belief_state, action, oj)
                distance = max([self.norm1(belief_state2, new_belief) for belief_state2 in self.belief_points])
                if distance > largest_distance:
                    largest_distance = distance
                    add_belief = new_belief
            self.belief_points = np.vstack((self.belief_points, add_belief))
         

def norm1(self, a, b):
    new = np.array(a) - np.array(b)
    distance = 0.0
    for i in new:
        distance += abs(i)
    return distance