def test_simple_asymm_ghost(use_discrete):
# Make opponent for asymmetric case
brain_name_opp = BRAIN_NAME + "Opp"
env = SimpleEnvironment(
[BRAIN_NAME + "?team=0", brain_name_opp + "?team=1"],
use_discrete=use_discrete)
self_play_settings = SelfPlaySettings(
play_against_latest_model_ratio=1.0,
save_steps=10000,
swap_steps=10000,
team_change=400,
)
config = attr.evolve(
PPO_TF_CONFIG,
self_play=self_play_settings,
max_steps=4000,
framework=FrameworkType.TENSORFLOW,
)
_check_environment_trains(env, {
BRAIN_NAME: config,
brain_name_opp: config
})
def test_simple_ghost_fails(use_discrete):
env = SimpleEnvironment([BRAIN_NAME + "?team=0", BRAIN_NAME + "?team=1"],
use_discrete=use_discrete)
# This config should fail because the ghosted policy is never swapped with a competent policy.
# Swap occurs after max step is reached.
self_play_settings = SelfPlaySettings(play_against_latest_model_ratio=1.0,
save_steps=2000,
swap_steps=4000)
config = attr.evolve(
PPO_TF_CONFIG,
self_play=self_play_settings,
max_steps=2500,
framework=FrameworkType.TENSORFLOW,
)
_check_environment_trains(env, {BRAIN_NAME: config},
success_threshold=None)
processed_rewards = [
default_reward_processor(rewards)
for rewards in env.final_rewards.values()
]
success_threshold = 0.9
assert any(reward > success_threshold
for reward in processed_rewards) and any(
reward < success_threshold for reward in processed_rewards)
def dummy_config():
return TrainerSettings(self_play=SelfPlaySettings())
def dummy_config():
return TrainerSettings(self_play=SelfPlaySettings(),
framework=FrameworkType.PYTORCH)