Skip to content

Upgrade to use RLLib master (almost v1.3). #10

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Manuscrit wants to merge 15 commits into
base: master
Choose a base branch
from
Open

Upgrade to use RLLib master (almost v1.3). #10

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
9f8a1b2
Upgrade to use RLLib master (almost v1.3).
Apr 15, 2021
4e0bf57
Use _compute_actions_helper instead of compute_actions in policies. S…
Apr 15, 2021
9888b75
Fixing some tests, _comute_action_helper, update_target
Apr 15, 2021
5599bf6
Fix amTFT exploiter.
Apr 16, 2021
00927ab
Support LSTM in amTFT.
Apr 20, 2021
07c35cc
Add meta game exp with LOLA-Exact.
Apr 27, 2021
c43019c
Add several options to stabilize the LOLA-PG training.
Apr 27, 2021
c1051f0
Fix bug in cross_play evaluator.
Apr 28, 2021
67c8f79
Add SSDMMCG vectorized.
Apr 28, 2021
6f8f7e4
Fix amTFT: how the environment is used during the rollouts
May 7, 2021
9dd9738
Change LOLA-Exact to make it work with discrete actions space larger …
May 7, 2021
028be50
Add stuff to the environments:
May 7, 2021
46ff4b0
Add meta game experiments with various meta solvers (like: alpha-rank…
May 7, 2021
c65be87
Add more meta algorithms in meta game experiment.
May 13, 2021
92ae8b4
Adapt LOLA-Exact and SOS-Exact to work with matrix games with N actio…
May 13, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions .gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,3 +50,4 @@ marltoolbox/examples/Tutorial_*.py
marltoolbox/experiments/Tutorial_*.py
api_key_wandb
wandb/
requirements.txt
18 changes: 18 additions & 0 deletions marltoolbox/algos/amTFT/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,3 +19,21 @@
AmTFTRolloutsTorchPolicy,
)
from marltoolbox.algos.amTFT.train_helper import train_amtft


__all__ = [
"train_amtft",
"AmTFTRolloutsTorchPolicy",
"amTFTQValuesTorchPolicy",
"Level1amTFTExploiterTorchPolicy",
"observation_fn",
"AmTFTCallbacks",
"WORKING_STATES",
"WORKING_STATES_IN_EVALUATION",
"AmTFTReferenceClass",
"DEFAULT_NESTED_POLICY_COOP",
"DEFAULT_NESTED_POLICY_SELFISH",
"PLOT_ASSEMBLAGE_TAGS",
"PLOT_KEYS",
"DEFAULT_CONFIG",
]
60 changes: 35 additions & 25 deletions marltoolbox/algos/amTFT/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,20 +4,23 @@
from ray.rllib.utils import merge_dicts

from marltoolbox.algos import hierarchical, augmented_dqn
from marltoolbox.utils import \
postprocessing, miscellaneous
from marltoolbox.utils import postprocessing, miscellaneous

logger = logging.getLogger(__name__)

APPROXIMATION_METHOD_Q_VALUE = "amTFT_use_Q_net"
APPROXIMATION_METHOD_ROLLOUTS = "amTFT_use_rollout"
APPROXIMATION_METHODS = (APPROXIMATION_METHOD_Q_VALUE,
APPROXIMATION_METHOD_ROLLOUTS)
WORKING_STATES = ("train_coop",
"train_selfish",
"eval_amtft",
"eval_naive_selfish",
"eval_naive_coop")
APPROXIMATION_METHODS = (
APPROXIMATION_METHOD_Q_VALUE,
APPROXIMATION_METHOD_ROLLOUTS,
)
WORKING_STATES = (
"train_coop",
"train_selfish",
"eval_amtft",
"eval_naive_selfish",
"eval_naive_coop",
)
WORKING_STATES_IN_EVALUATION = WORKING_STATES[2:]

OWN_COOP_POLICY_IDX = 0
Expand All @@ -29,8 +32,9 @@
DEFAULT_NESTED_POLICY_COOP = DEFAULT_NESTED_POLICY_SELFISH.with_updates(
postprocess_fn=miscellaneous.merge_policy_postprocessing_fn(
postprocessing.welfares_postprocessing_fn(
add_utilitarian_welfare=True, ),
postprocess_nstep_and_prio
add_utilitarian_welfare=True,
),
postprocess_nstep_and_prio,
)
)

Expand All @@ -44,39 +48,44 @@
"rollout_length": 40,
"n_rollout_replicas": 20,
"last_k": 1,
"punish_instead_of_selfish": False,
# TODO use log level of RLLib instead of mine
"verbose": 1,
"auto_load_checkpoint": True,

# To configure
"own_policy_id": None,
"opp_policy_id": None,
"callbacks": None,
# One from marltoolbox.utils.postprocessing.WELFARES
"welfare_key": postprocessing.WELFARE_UTILITARIAN,

'nested_policies': [
"nested_policies": [
# Here the trainer need to be a DQNTrainer
# to provide the config for the 3 DQNTorchPolicy
{"Policy_class": DEFAULT_NESTED_POLICY_COOP,
"config_update": {}},
{"Policy_class": DEFAULT_NESTED_POLICY_SELFISH,
"config_update": {}},
{"Policy_class": DEFAULT_NESTED_POLICY_COOP,
"config_update": {}},
{"Policy_class": DEFAULT_NESTED_POLICY_SELFISH,
"config_update": {}},
{"Policy_class": DEFAULT_NESTED_POLICY_COOP, "config_update": {}},
{
"Policy_class": DEFAULT_NESTED_POLICY_SELFISH,
"config_update": {},
},
{"Policy_class": DEFAULT_NESTED_POLICY_COOP, "config_update": {}},
{
"Policy_class": DEFAULT_NESTED_POLICY_SELFISH,
"config_update": {},
},
],
"optimizer": {"sgd_momentum": 0.0, },
}
"optimizer": {
"sgd_momentum": 0.0,
},
"batch_mode": "complete_episodes",
},
)

PLOT_KEYS = [
"punish",
"debit",
"debit_threshold",
"summed_debit",
"summed_n_steps_to_punish"
"summed_n_steps_to_punish",
"reset_rnn_state",
]

PLOT_ASSEMBLAGE_TAGS = [
Expand All @@ -85,6 +94,7 @@
("debit_threshold",),
("summed_debit",),
("summed_n_steps_to_punish",),
("reset_rnn_state",),
]


Expand Down
Loading