act is called at every time step of the game:
def act(self, game_state):
...
return action
where action is one of the following 6 strings
["UP", "DOWN", "LEFT", "RIGHT", "WAIT", "BOMB"]
This is used to test behaviours independently.
behave at every step of the game by the genetic agent:
def behave(self, game_state):
...
return {action: score,...}
is the dict accessible by act, it contains the whole observable game state as described here:
state = {
'round': self.round,
'step': self.step,
'field': np.array(self.arena),
'self': agent.get_state(),
'others': [other.get_state() for other in self.active_agents if other is not agent],
'bombs': [bomb.get_state() for bomb in self.bombs],
'coins': [coin.get_state() for coin in self.coins if coin.collectable],
'user_input': self.user_input,
'dead_zones': <function>(coord,t)
}
Note: now the game state provides under "dead_zones" function to evaluate immediately deadly positions: the function requires two arguments: the coords (x,y) and the time (t). The function returns True when deadly.
| key | type | shape | values | meaning |
|---|---|---|---|---|
| round | int | |||
| step | int | |||
| field | array | [17,17] | -1, 0, 1 | wall , free , crate |
| self | array | [4] | [str,int,bool,(int,int)] | [name, score, bombs_left, (x,y)] |
| others | array | [n] | [[str,int,bool,(int,int)],...] | [[name, score, bombs_left, (x,y)],...] |
| bombs | array | [n] | [[int,int],...] | [[x,y],...] |
| coins | array | [n] | [[int,int],...] | [[x,y],...] |
| user_input | ? | ? | ? | ? |
| explosion_map | array | [17,17] | 0,1 | safe,explosion |
To run a simulation the best approach is to use the command:
python main.py play --match-name "wall_breaker_t1" --save-replay --n-rounds 10 --my-agent wall_breaker
python main.py play --n-rounds 10 --my-agent genetic_agent --train_genetic
python main.py play --n-rounds 10 --my-agent genetic_agent
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent