def update_particles(particles, action_policy, observation):
    (particles, particle_weights) = particles
    new_particles = []
    new_particle_weights = {}
    n = len(particles)
    for i in range(n):
        # sample() will pick a key from a dictionary,
        # with probablity := value of that key
        j = sample(particle_weights)
        # action_policy() will return next state of agent, given current state
        # Note that action_policy() itself can be stochastic.
        j = action_policy(j)
        # observation() gives probability of state, using evidence
        # An example of evidence is sensor data of a robot. 
        new_particle_weights[j] = observation(j)
        new_particles.append(j)
    
    # normalize(dict) will divide all values in dict
    # by sum([v for k,v in dict.items()]) 
    normalize(new_particle_weights)

    return (new_particles,new_particle_weights)