def dict_to_flatten_board_state(state, map_object):
""" Converts a game state to its flatten (list) board state representation.
:param state: A game state.
:param map_object: The instantiated Map object
:return: A flatten (list) representation of the phase (81*35 = 2835)
"""
state_proto = dict_to_proto(state, StateProto)
return proto_to_board_state(state_proto, map_object).flatten().tolist()
def dict_to_flatten_prev_orders_state(phase, map_object):
""" Converts a phase to its flatten (list) prev orders state representation.
:param phase: A phase from a saved game.
:param map_object: The instantiated Map object
:return: A flatten (list) representation of the prev orders (81*40 = 3240)
"""
phase_proto = dict_to_proto(phase, PhaseHistoryProto)
return proto_to_prev_orders_state(phase_proto,
map_object).flatten().tolist()
def extract_state_proto(game):
""" Extracts the state_proto from a diplomacy.Game object
:type game: diplomacy.Game
"""
state = game.get_state()
state['game_id'] = game.game_id
state['map'] = game.map.name
state['rules'] = list(game.rules)
return dict_to_proto(state, StateProto)
def process_game(line):
""" Process a line in the .jsonl file
:return: A tuple (game_id, saved_game_zlib)
"""
if not line:
return None, None
saved_game = json.loads(line)
saved_game = add_cached_states_to_saved_game(saved_game)
saved_game = add_possible_orders_to_saved_game(saved_game)
saved_game_proto = dict_to_proto(saved_game, SavedGameProto)
saved_game_proto = add_rewards_to_saved_game_proto(saved_game_proto, DefaultRewardFunction())
saved_game_zlib = proto_to_zlib(saved_game_proto)
return saved_game['id'], saved_game_zlib
def extract_phase_history_proto(game,
nb_previous_phases=NB_PREV_ORDERS_HISTORY):
""" Extracts the phase_history_proto from a diplomacy.Game object
:param game: The diplomacy.Game object
:param nb_previous_phases: Integer. If set, only the last x phases will be returned.
If None, the full history since the beginning of the game is returned.
:return: A list of `.proto.game.PhaseHistory` proto.
:type game: diplomacy.Game
"""
from_phase = None if nb_previous_phases is None else -1 * nb_previous_phases
phase_history = Game.get_phase_history(game, from_phase=from_phase)
return [
dict_to_proto(hist.to_dict(), PhaseHistoryProto)
for hist in phase_history
]
def get_saved_game_proto(env, players, stored_board_state,
stored_prev_orders_state, stored_possible_orders,
power_variables, start_phase_ix, reward_fn,
advantage_fn, is_partial_game):
""" Extracts the saved game proto from the environment to send back to the learner
:param env: The gym environment (needs to implement a SaveGame wrapper)
:param players: A list of instantiated players
:param stored_board_state: A dictionary with phase_name as key and board_state as value
:param stored_prev_orders_state: A dictionary with phase_name as key and prev_orders_state as value
:param stored_possible_orders: A dictionary with phase_name as key and possible orders as value
:param power_variables: A dict containing orders, policy details, values, rewards, returns for each power
:param start_phase_ix: For partial game, the index of the phase from which to start learning
:param reward_fn: The reward function to use to calculate rewards
:param advantage_fn: An instance of `.models.self_play.advantages`
:param is_partial_game: Boolean that indicates that we are processing an incomplete game
:return: The saved game in proto format
"""
# pylint: disable=too-many-arguments
powers = sorted(
[power_name for power_name in get_map_powers(env.game.map)])
assigned_powers = env.get_all_powers_name()
# Computing returns
for power_name in powers:
rewards = power_variables[power_name]['rewards']
state_values = power_variables[power_name]['state_values']
last_state_value = power_variables[power_name]['last_state_value']
power_variables[power_name]['returns'] = advantage_fn.get_returns(
rewards, state_values, last_state_value)
# Retrieving saved game
saved_game = env.get_saved_game()
# Restoring stored variables on the saved game before converting to proto
for phase_ix, phase in enumerate(saved_game['phases']):
# Last phase - Only storing state value
if phase_ix == len(saved_game['phases']) - 1:
state_values = {
power_name:
float(power_variables[power_name]['state_values'][-1])
for power_name in powers
}
phase['state_value'] = state_values
break
# Setting shared fields (board_state, prev_orders_state, possible_orders)
phase['state']['board_state'] = stored_board_state[phase['name']]
if phase['name'][-1] == 'M':
phase['prev_orders_state'] = stored_prev_orders_state[
phase['name']]
phase['possible_orders'] = {
loc: stored_possible_orders[phase['name']][loc].value
for loc in stored_possible_orders[phase['name']]
}
# Setting orders, policy_details, state_values
phase['orders'] = {
power_name: power_variables[power_name]['orders'][phase_ix]
for power_name in powers
}
phase['policy'] = {
power_name: power_variables[power_name]['policy_details'][phase_ix]
for power_name in powers
}
phase['state_value'] = {
power_name:
float(power_variables[power_name]['state_values'][phase_ix])
for power_name in powers
}
# Adding power assignments, done reason, and kwargs
done_reason = env.done_reason.value if env.done_reason is not None else ''
saved_game['done_reason'] = done_reason
saved_game['assigned_powers'] = assigned_powers
saved_game['players'] = [player.name for player in players]
saved_game['kwargs'] = {
power_name: players[assigned_powers.index(power_name)].kwargs
for power_name in powers
}
saved_game['is_partial_game'] = is_partial_game
saved_game['start_phase_ix'] = start_phase_ix if is_partial_game else 0
saved_game['reward_fn'] = reward_fn.name
saved_game['rewards'] = {
power_name: power_variables[power_name]['rewards']
for power_name in powers
}
saved_game['returns'] = {
power_name: power_variables[power_name]['returns']
for power_name in powers
}
# Returning
return dict_to_proto(saved_game, SavedGameProto)
def extract_possible_orders_proto(game):
""" Extracts the possible_orders_proto from a diplomacy.Game object
:type game: diplomacy.Game
"""
possible_orders = game.get_all_possible_orders()
return dict_to_proto(possible_orders, MapStringList)
def test_to_from_dict():
""" Tests proto_to_dict and dict_to_proto """
message_proto = _get_message()
message_dict = proto_to_dict(message_proto)
new_message_proto = dict_to_proto(message_dict, Message)
_compare_messages(message_proto, new_message_proto)