dodox/Columbus
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fixed Off-By-One for final reward

Browse Source
This commit is contained in:
Dominik Moritz Roth 2022-10-11 18:51:45 +02:00
parent d4a1a35190
commit e8c3b5dac0
1 changed files with 6 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
columbus/env.py
View File

@ -78,8 +78,7 @@ class ColumbusEnv(gym.Env):
self.aux_penalty_max = aux_penalty_max # 0 = off self.aux_penalty_max = aux_penalty_max # 0 = off
self.aux_reward_discretize = aux_reward_discretize self.aux_reward_discretize = aux_reward_discretize
# 0 = dont discretize; how many steps (along diagonal) # 0 = dont discretize; how many steps (along diagonal)
self.aux_reward_discretize = 0 self.penalty_from_edges = False # not ready yet...
self.penalty_from_edges = False
self.draw_observable = True self.draw_observable = True
self.draw_joystick = True self.draw_joystick = True
self.draw_entities = True self.draw_entities = True
@ -105,6 +104,7 @@ class ColumbusEnv(gym.Env):
self._seed(self.env_seed) self._seed(self.env_seed)
self._init = False self._init = False
self._term_next = False
@property @property
def observation_space(self): def observation_space(self):
@ -177,7 +177,7 @@ class ColumbusEnv(gym.Env):
if self.penalty_from_edges: if self.penalty_from_edges:
penalty = self.aux_penalty_max / \ penalty = self.aux_penalty_max / \
(1 + self.sq_dist(entity.pos, (1 + self.sq_dist(entity.pos,
self.agent.pos) - entity.radius - self.agent.redius) self.agent.pos) - entity.radius - self.agent.radius)
else: else:
penalty = self.aux_penalty_max / \ penalty = self.aux_penalty_max / \
(1 + self.sq_dist(entity.pos, self.agent.pos)) (1 + self.sq_dist(entity.pos, self.agent.pos))
@ -214,8 +214,9 @@ class ColumbusEnv(gym.Env):
self.score += reward # aux_reward does not count towards the score self.score += reward # aux_reward does not count towards the score
if self.aux_reward_max: if self.aux_reward_max:
reward += self._get_aux_reward() reward += self._get_aux_reward()
done = self.die_on_zero and self.score <= 0 or self.return_on_score != - \ done = self._term_next or (self.die_on_zero and self.score <= 0 or self.return_on_score != -
1 and self.score > self.return_on_score or self.terminate_on_reward and gotRew 1 and self.score > self.return_on_score)
self._term_next = self.terminate_on_reward and gotRew
info = {'score': self.score, 'reward': reward} info = {'score': self.score, 'reward': reward}
self._rendered = False self._rendered = False
if done: if done: