gistfile1.py

def choose_actions(sensors, predicted_tracks, reward_function, smapping=[0,2], tmapping=[0,2]):
    """Work out which action to take. Sensors must be able to return actions, a collection which is either
    iterable or testable for membership. Tracks have been predicted to a time of interest in advance (this 
    logic could exist inside choose actions with the addition of a :class:`Predictor`. A reward function is
    passed - or could be accessed from the parent state. Mappings to the x,y position of sensor and track are
    provided - it's assumed that the angle to target is in x,y on a flat earth.)"""
    
    # Firstly, can a sensor see a target? Create a dictionary of those 
    sensor_target_pair = dict()
    for sensor in sensors:
        sensor_target_pair[sensor] = set()
        for track in predicted_tracks:
            relative_position = track.state_vector[tmapping] - sensor.position[smapping]
            angle_to_target = np.arctan2(relative_position[1], relative_position[0])
            
            if angle_to_target in sensor.actions(predicted_track.timestamp):
                sensor_target_pair[sensor].add(track)
                           
    # From this create the list of possible configurations
    configs = ({sensor: action 
           for sensor, action in zip(sensor_target_pair.keys(), config)}
           for config in itertools.product(*sensor_target_pair.values()))
    
    best_reward = -inf
    selected_config = None
    for config in configs:
        reward = reward_function(config)
        if reward > best_reward:
            selected_config = config
            best_reward = reward
        
    return config, best_reward