File size: 2,137 Bytes
97b6013 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 |
#-*-Python-*-
create_maze_env.env_name = "AntMaze"
context_range = (%CONTEXT_RANGE_MIN, %CONTEXT_RANGE_MAX)
meta_context_range = ((-4, -4), (20, 20))
RESET_EPISODE_PERIOD = 500
RESET_ENV_PERIOD = 1
# End episode every N steps
UvfAgent.reset_episode_cond_fn = @every_n_steps
every_n_steps.n = %RESET_EPISODE_PERIOD
train_uvf.max_steps_per_episode = %RESET_EPISODE_PERIOD
# Do a manual reset every N episodes
UvfAgent.reset_env_cond_fn = @every_n_episodes
every_n_episodes.n = %RESET_ENV_PERIOD
every_n_episodes.steps_per_episode = %RESET_EPISODE_PERIOD
## Config defaults
EVAL_MODES = ["eval1", "eval2", "eval3"]
## Config agent
CONTEXT = @agent/Context
META_CONTEXT = @meta/Context
## Config agent context
agent/Context.context_ranges = [%context_range]
agent/Context.context_shapes = [%SUBGOAL_DIM]
agent/Context.meta_action_every_n = 10
agent/Context.samplers = {
"train": [@train/DirectionSampler],
"explore": [@train/DirectionSampler],
}
agent/Context.context_transition_fn = @relative_context_transition_fn
agent/Context.context_multi_transition_fn = @relative_context_multi_transition_fn
agent/Context.reward_fn = @uvf/negative_distance
## Config meta context
meta/Context.context_ranges = [%meta_context_range]
meta/Context.context_shapes = [2]
meta/Context.samplers = {
"train": [@train/RandomSampler],
"explore": [@train/RandomSampler],
"eval1": [@eval1/ConstantSampler],
"eval2": [@eval2/ConstantSampler],
"eval3": [@eval3/ConstantSampler],
}
meta/Context.reward_fn = @task/negative_distance
## Config rewards
task/negative_distance.state_indices = [0, 1]
task/negative_distance.relative_context = False
task/negative_distance.diff = False
task/negative_distance.offset = 0.0
## Config samplers
train/RandomSampler.context_range = %meta_context_range
train/DirectionSampler.context_range = %context_range
train/DirectionSampler.k = %SUBGOAL_DIM
relative_context_transition_fn.k = %SUBGOAL_DIM
relative_context_multi_transition_fn.k = %SUBGOAL_DIM
MetaAgent.k = %SUBGOAL_DIM
eval1/ConstantSampler.value = [16, 0]
eval2/ConstantSampler.value = [16, 16]
eval3/ConstantSampler.value = [0, 16]
|