def play_best(game: Game, save_game: bool,agent_nation: list, turn_number = 3, repeat_number= 1000):
set_starting_influence(game)
if save_game:
saver = GameSaver()
q_table_Handler = QtableHandler(game, agent_nation)
iterator = 0
state = 0
while not game.is_game_done:
iterator += 1
q_table_Handler.set_turn_info()
# settings order
for power_name, power in game.powers.items():
if power_name in agent_nation:
power_orders = q_table_Handler.chose_best(power_name)
else:
power_orders = q_table_Handler.chose_orders(power_name)
game.set_orders(power_name, power_orders)
if save_game:
saver.save_game(game, "gierka")
game.process()
adjust_influence(game)
if iterator == turn_number:
return
def add_possible_orders_to_saved_game(saved_game):
""" Adds possible_orders for each phase of the saved game """
if saved_game['map'].startswith('standard'):
for phase in saved_game['phases']:
game = Game(map_name=saved_game['map'], rules=saved_game['rules'])
game.set_state(phase['state'])
phase['possible_orders'] = game.get_all_possible_orders()
return saved_game
def test_mis_003():
""" Tests the MIS response """
daide_str = 'MIS ( #0 )'
game = Game(map_name='standard')
phase_name = 'W1901A'
power = game.get_power('FRANCE')
response = responses.MIS(phase_name=phase_name, power=power)
assert isinstance(response, responses.MIS), 'Expected a MIS response'
assert bytes(response) == str_to_bytes(daide_str)
def test_mis_010():
""" Tests the MIS response """
daide_str = 'MIS ( FRA FLT BRE ) ( FRA AMY MAR )'
game = Game(map_name='standard')
phase_name = 'S1901M'
power = game.get_power('FRANCE')
power.orders['INVALID'] = 'A PAR - BUR'
response = responses.MIS(phase_name=phase_name, power=power)
assert isinstance(response, responses.MIS), 'Expected a MIS response'
assert bytes(response) == str_to_bytes(daide_str)
def test_get_nb_centers():
""" Testing if the number of supply centers is correct """
game = Game()
player = FakePlayer()
state_proto = extract_state_proto(game)
# Checking every power
power_names = [power_name for power_name in game.powers]
for power_name in power_names:
assert player.get_nb_centers(state_proto, power_name) == len(game.get_power(power_name).centers)
def test_mis_012():
""" Tests the MIS response """
daide_str = 'MIS ( #-1 )'
game = Game(map_name='standard')
phase_name = 'W1901A'
power = game.get_power('FRANCE')
power.centers += ['LON']
power.units = power.units[:2]
response = responses.MIS(phase_name=phase_name, power=power)
assert isinstance(response, responses.MIS), 'Expected a MIS response'
assert bytes(response) == str_to_bytes(daide_str)
def test_mis_002():
""" Tests the MIS response """
daide_str = 'MIS ( TUR FLT ANK MRT ( ARM ) ) ' \
'( TUR FLT CON MRT ( BLA SMY ( BUL ECS ) ( BUL SCS ) ) ) ' \
'( TUR AMY SMY MRT ( SYR ) )'
game = Game(map_name='standard')
phase_name = 'S1901R'
power = game.get_power('TURKEY')
power.units = ['F ANK', 'F CON', 'A SMY']
power.retreats['F ANK'] = ['ARM']
power.retreats['F CON'] = ['BLA', 'SMY', 'BUL/EC', 'BUL/SC']
power.retreats['A SMY'] = ['SYR']
response = responses.MIS(phase_name=phase_name, power=power)
assert isinstance(response, responses.MIS), 'Expected a MIS response'
assert bytes(response) == str_to_bytes(daide_str)
def test_get_state_value(self):
""" Checks if the .get_state_value method works """
game = Game()
state_proto = extract_state_proto(game)
phase_history_proto = extract_phase_history_proto(game)
kwargs = {
'player_seed': 0,
'noise': 0.,
'temperature': 0.,
'dropout_rate': 0.
}
# Testing with and without prefetching
for use_prefetching in (False, True):
if not use_prefetching:
results = yield self.adapter.get_state_value(
state_proto, 'FRANCE', phase_history_proto, **kwargs)
else:
fetches = yield self.adapter.get_state_value(
state_proto,
'FRANCE',
phase_history_proto,
prefetch=True,
**kwargs)
fetches = yield process_fetches_dict(self.queue_dataset,
fetches)
results = yield self.adapter.get_state_value(
state_proto,
'FRANCE',
phase_history_proto,
fetches=fetches,
**kwargs)
assert results != 0.
def test_board_state():
""" Tests the proto_to_state_space """
game = Game()
game_map = game.map
state_proto = state_space.extract_state_proto(game)
new_game = state_space.build_game_from_state_proto(state_proto)
# Retrieving board_state
state_proto_2 = state_space.extract_state_proto(new_game)
board_state_1 = state_space.proto_to_board_state(state_proto, game_map)
board_state_2 = state_space.proto_to_board_state(state_proto_2, game_map)
# Checking
assert np.allclose(board_state_1, board_state_2)
assert board_state_1.shape == (state_space.NB_NODES, state_space.NB_FEATURES)
assert game.get_hash() == new_game.get_hash()
def get_opening_orders(self):
""" Returns a dictionary of power_name: [orders] for each power
The orders represent the opening orders that would have been submitted by the player
"""
game = Game()
state_proto = extract_state_proto(game)
phase_history_proto = extract_phase_history_proto(game)
possible_orders_proto = extract_possible_orders_proto(game)
# Retrieving all orders
# Using default player_seed, noise, temperature, and dropout_rate.
# power_orders is a list of tuples (orders, policy_details)
power_orders = yield [
self.policy_adapter.get_orders(self.get_orderable_locations(
state_proto, power_name),
state_proto,
power_name,
phase_history_proto,
possible_orders_proto,
retry_on_failure=False)
for power_name in game.powers
]
return {
power_name: orders[0]
for power_name, orders in zip(game.powers.keys(), power_orders)
}
def test_now():
""" Tests the NOW response """
daide_str = 'NOW ( SPR #1901 ) ( AUS AMY BUD ) ( AUS AMY VIE ) ( AUS FLT TRI ) ( ENG FLT EDI )' \
' ( ENG FLT LON ) ( ENG AMY LVP ) ( FRA FLT BRE ) ( FRA AMY MAR ) ( FRA AMY PAR )' \
' ( GER FLT KIE ) ( GER AMY BER ) ( GER AMY MUN ) ( ITA FLT NAP ) ( ITA AMY ROM )' \
' ( ITA AMY VEN ) ( RUS AMY WAR ) ( RUS AMY MOS ) ( RUS FLT SEV )' \
' ( RUS FLT ( STP SCS ) ) ( TUR FLT ANK ) ( TUR AMY CON ) ( TUR AMY SMY )'
game = Game(map_name='standard')
phase_name = game.get_current_phase()
units = {power.name: power.units for power in game.powers.values()}
retreats = {power.name: power.retreats for power in game.powers.values()}
response = responses.NOW(phase_name=phase_name,
powers_units=units,
powers_retreats=retreats)
assert isinstance(response, responses.NOW), 'Expected a NOW response'
assert bytes(response) == str_to_bytes(daide_str)
def get_player_seeds(self):
""" Returns a dictionary of power_name: seed to use for all powers """
map_object = (self.game or Game()).map
if not self.game:
return {power_name: 0 for power_name in get_map_powers(map_object)}
return {
power_name: get_player_seed(self.game.game_id, power_name)
for power_name in get_map_powers(map_object)
}
def test_ord_004():
""" Tests the ORD response """
daide_str = 'ORD ( SPR #1901 ) ( ( ENG FLT NWG ) SUP ( ENG AMY YOR ) MTO NWY ) ( NSO )'
order_daide_str = '( ENG FLT NWG ) SUP ( ENG AMY YOR ) MTO NWY'
game = Game(map_name='standard')
phase_name = game.map.phase_abbr(game.phase)
response = responses.ORD(phase_name=phase_name,
order_bytes=str_to_bytes(order_daide_str),
results=[BOUNCE.code, DISLODGED])
assert isinstance(response, responses.ORD), 'Expected a ORD response'
assert bytes(response) == str_to_bytes(daide_str)
def test_get_orderable_locations():
""" Testing if the number of orderable locations is correct """
game = Game()
player = FakePlayer()
state_proto = extract_state_proto(game)
# Checking every power
power_names = [power_name for power_name in game.powers]
for power_name in power_names:
expected_locs = [unit.replace('*', '')[2:5] for unit in state_proto.units[power_name].value]
expected_locs += state_proto.builds[power_name].homes
assert sorted(player.get_orderable_locations(state_proto, power_name)) == sorted(expected_locs)
def play_game(game: Game, save_game: bool, agent_nation: list, label="", turn_number = 3, repeat_number= 1000):
set_starting_influence(game)
if save_game:
saver = GameSaver()
q_table_Handler = QtableHandler(game, agent_nation)
iterator = 0
state = 0
finish = False
stats = {"centers":defaultdict(list), "influence":defaultdict(list)}
while not game.is_game_done and not finish:
iterator += 1
q_table_Handler.set_turn_info()
# settings order
phase = game.get_current_phase()[-1]
for power_name, power in game.powers.items():
power_orders = q_table_Handler.chose_orders(power_name)
game.set_orders(power_name, power_orders)
#visualizer.paint_orders(game)
if save_game:
saver.save_game(game, "gierka")
game.process()
if phase == 'M':
q_table_Handler.set_reward()
adjust_influence(game)
if iterator == turn_number:
state += 1
if state % repeat_number == 0:
q_table_Handler.save()
if state == repeat_number:
save_stat(stats,turn_number,label)
game = load_saved_games_from_disk("game.json")[0]
return
iterator = 0
print("State: ", state)
print("Accuracy: {0}".format(q_table_Handler.get_accuracy()))
print("Number of Germany centers: ", game.get_centers("GERMANY").__len__(), game.get_power("GERMANY").influence.__len__())
for power_name, _ in game.powers.items():
stats["influence"][power_name].append(game.get_power(power_name).influence.__len__())
stats["centers"][power_name].append(game.get_centers(power_name).__len__())
game = load_saved_games_from_disk("game.json")[0]
q_table_Handler.game = game
q_table_Handler.attempts = 0
q_table_Handler.miss_hits = 0
def test_mis_008():
""" Tests the MIS response """
daide_str = 'MIS ( FRA FLT BRE ) ( FRA AMY MAR )'
game = Game(map_name='standard')
game.add_rule('NO_CHECK')
game.set_orders('FRANCE', ['A PAR - BUR'])
phase_name = 'S1901M'
power = game.get_power('FRANCE')
response = responses.MIS(phase_name=phase_name, power=power)
assert isinstance(response, responses.MIS), 'Expected a MIS response'
assert bytes(response) == str_to_bytes(daide_str)
def test_sco():
""" Tests the SCO response """
daide_str = 'SCO ( AUS BUD TRI VIE ) ( ENG EDI LON LVP ) ( FRA BRE MAR PAR ) ' \
'( GER BER KIE MUN ) ( ITA NAP ROM VEN ) ( RUS MOS SEV STP WAR ) ' \
'( TUR ANK CON SMY ) ( UNO BEL BUL DEN GRE HOL NWY POR RUM SER SPA SWE TUN )'
game = Game(map_name='standard')
power_centers = {
power.name: power.centers
for power in game.powers.values()
}
response = responses.SCO(power_centers, map_name='standard')
assert isinstance(response, responses.SCO), 'Expected a SCO response'
assert bytes(response) == str_to_bytes(daide_str)
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 test_int_norm_centers_reward():
""" Tests for InterimNormNbCentersReward """
game = Game()
rew_fn = IntNormNbCentersReward()
# Removing one center from FRANCE and adding it to GERMANY
prev_state_proto = extract_state_proto(game)
for power in game.powers.values():
if power.name == 'FRANCE':
power.centers.remove('PAR')
if power.name == 'GERMANY':
power.centers.append('PAR')
state_proto = extract_state_proto(game)
assert rew_fn.name == 'int_norm_nb_centers_reward'
get_reward = lambda power_name, is_terminal, done_reason: rew_fn.get_reward(
prev_state_proto,
state_proto,
power_name,
is_terminal_state=is_terminal,
done_reason=done_reason)
# --- Not in terminal state
assert get_reward('AUSTRIA', False, None) == 0.
assert get_reward('ENGLAND', False, None) == 0.
assert get_reward('FRANCE', False, None) == -1. / 18
assert get_reward('GERMANY', False, None) == 1. / 18
assert get_reward('ITALY', False, None) == 0.
assert get_reward('RUSSIA', False, None) == 0.
assert get_reward('TURKEY', False, None) == 0.
# --- In terminal state
assert get_reward('AUSTRIA', True, DoneReason.GAME_ENGINE) == 0.
assert get_reward('ENGLAND', True, DoneReason.GAME_ENGINE) == 0.
assert get_reward('FRANCE', True, DoneReason.GAME_ENGINE) == -1. / 18
assert get_reward('GERMANY', True, DoneReason.GAME_ENGINE) == 1. / 18
assert get_reward('ITALY', True, DoneReason.GAME_ENGINE) == 0.
assert get_reward('RUSSIA', True, DoneReason.GAME_ENGINE) == 0.
assert get_reward('TURKEY', True, DoneReason.GAME_ENGINE) == 0.
# --- Thrashing
assert get_reward('AUSTRIA', True, DoneReason.THRASHED) == -1.
assert get_reward('ENGLAND', True, DoneReason.THRASHED) == -1.
assert get_reward('FRANCE', True, DoneReason.THRASHED) == -1.
assert get_reward('GERMANY', True, DoneReason.THRASHED) == -1.
assert get_reward('ITALY', True, DoneReason.THRASHED) == -1.
assert get_reward('RUSSIA', True, DoneReason.THRASHED) == -1.
assert get_reward('TURKEY', True, DoneReason.THRASHED) == -1.
def main():
try:
(options, arguments) = getopt.getopt(sys.argv[1:], 'h')
except getopt.error:
sys.exit("Unknown input parameter.")
if [] == arguments:
arguments = ["shiftLeft"]
if not os.path.exists(arguments[0] + ".map"):
sys.exit("%s.map could not be opened" % (arguments[0], ))
game = Game(map_name=arguments[0])
while not game.is_game_done:
# Getting the list of possible orders for all locations
possible_orders = game.get_all_possible_orders()
# For each power, randomly sampling a valid order
for power_name, power in game.powers.items():
# power_orders = [random.choice(possible_orders[loc]) for loc in game.get_orderable_locations(power_name)
# if possible_orders[loc]]
power_orders = []
for loc in game.get_orderable_locations(power_name):
if '/' == loc[-1]:
loc = loc[:-1]
if possible_orders[loc]:
power_orders.append(random.choice(possible_orders[loc]))
game.set_orders(power_name, power_orders)
# Messages can be sent locally with game.add_message
# e.g. game.add_message(Message(sender='FRANCE',
# recipient='ENGLAND',
# message='This is a message',
# phase=self.get_current_phase(),
# time_sent=int(time.time())))
# Processing the game to move to the next phase
game.process()
# to_saved_game_format(game, output_path='collected_autoplay_games.json')
# Exporting the game to disk to visualize (game is appended)
with open('collected_autoplay_games.json', 'a') as outp:
outp.write(to_saved_game_format(game))
def __init__(self, env, players, power_assignments):
""" Constructor
:param env: The wrapped env
:param players: A list of player instances
:param power_assignments: The list of powers (in order) to assign to each player
:type env: gym.core.Env
:type players: List[diplomacy_research.players.player.Player]
"""
super(AssignPlayers, self).__init__(env)
game = self.game or Game()
# Making sure we have the correct number of powers
assert len(power_assignments) == len(game.powers.keys())
assert sorted(power_assignments) == sorted(game.powers.keys())
# Setting fixed ordering
self._powers = power_assignments
self._players = players
def test_get_feedable_item(self):
""" Checks if the .get_feedable_item method works """
game = Game()
state_proto = extract_state_proto(game)
phase_history_proto = extract_phase_history_proto(game)
possible_orders_proto = extract_possible_orders_proto(game)
locs = ['PAR', 'MAR', 'BUR']
kwargs = {
'player_seed': 0,
'noise': 0.,
'temperature': 0.,
'dropout_rate': 0.
}
assert self.dataset_builder.get_feedable_item(locs, state_proto,
'FRANCE',
phase_history_proto,
possible_orders_proto,
**kwargs)
def check_serving(player, serving_port):
""" Makes sure the current serving process is still active, otherwise restarts it.
:param player: A player object to query the server
:param serving_port: The port to use for TF serving
"""
game = Game()
# Trying to check orders
for _ in range(MAX_SENTINEL_CHECKS):
orders = yield player.get_orders(game, 'FRANCE')
if orders:
return
# Could not get orders x times in a row, restarting process
LOGGER.warning(
'Could not retrieve orders from the serving process after %d attempts.',
MAX_SENTINEL_CHECKS)
LOGGER.warning('Restarting TF serving server.')
launch_serving(None, serving_port)
def test_get_draw_prob(self):
""" Checks if the .get_draw_prob method works """
game = Game()
state_proto = extract_state_proto(game)
phase_history_proto = extract_phase_history_proto(game)
possible_orders_proto = extract_possible_orders_proto(game)
locs = ['PAR', 'MAR', 'BUR']
kwargs = {
'player_seed': 0,
'noise': 0.,
'temperature': 1.,
'dropout_rate': 0.
}
# Temperature == 1.
# With and without prefetching
for use_prefetching in (False, True):
if not use_prefetching:
_, policy_details = yield self.adapter.get_orders(
locs, state_proto, 'FRANCE', phase_history_proto,
possible_orders_proto, **kwargs)
else:
fetches = yield self.adapter.get_orders(locs,
state_proto,
'FRANCE',
phase_history_proto,
possible_orders_proto,
prefetch=True,
**kwargs)
fetches = yield process_fetches_dict(self.queue_dataset,
fetches)
_, policy_details = yield self.adapter.get_orders(
locs,
state_proto,
'FRANCE',
phase_history_proto,
possible_orders_proto,
fetches=fetches,
**kwargs)
assert policy_details['draw_action'] in (True, False)
assert 0. < policy_details['draw_prob'] < 1.
def test_norm_centers_reward():
""" Tests for NormNbCentersReward """
game = Game()
rew_fn = NormNbCentersReward()
prev_state_proto = extract_state_proto(game)
state_proto = extract_state_proto(game)
assert rew_fn.name == 'norm_nb_centers_reward'
get_reward = lambda power_name, is_terminal, done_reason: rew_fn.get_reward(
prev_state_proto,
state_proto,
power_name,
is_terminal_state=is_terminal,
done_reason=done_reason)
# --- Not in terminal state
assert get_reward('AUSTRIA', False, None) == 0.
assert get_reward('ENGLAND', False, None) == 0.
assert get_reward('FRANCE', False, None) == 0.
assert get_reward('GERMANY', False, None) == 0.
assert get_reward('ITALY', False, None) == 0.
assert get_reward('RUSSIA', False, None) == 0.
assert get_reward('TURKEY', False, None) == 0.
# --- In terminal state
assert get_reward('AUSTRIA', True, DoneReason.GAME_ENGINE) == 3. / 18
assert get_reward('ENGLAND', True, DoneReason.GAME_ENGINE) == 3. / 18
assert get_reward('FRANCE', True, DoneReason.GAME_ENGINE) == 3. / 18
assert get_reward('GERMANY', True, DoneReason.GAME_ENGINE) == 3. / 18
assert get_reward('ITALY', True, DoneReason.GAME_ENGINE) == 3. / 18
assert get_reward('RUSSIA', True, DoneReason.GAME_ENGINE) == 4. / 18
assert get_reward('TURKEY', True, DoneReason.GAME_ENGINE) == 3. / 18
# --- Thrashing
assert get_reward('AUSTRIA', True, DoneReason.THRASHED) == 0.
assert get_reward('ENGLAND', True, DoneReason.THRASHED) == 0.
assert get_reward('FRANCE', True, DoneReason.THRASHED) == 0.
assert get_reward('GERMANY', True, DoneReason.THRASHED) == 0.
assert get_reward('ITALY', True, DoneReason.THRASHED) == 0.
assert get_reward('RUSSIA', True, DoneReason.THRASHED) == 0.
assert get_reward('TURKEY', True, DoneReason.THRASHED) == 0.
def get_opening_orders(self):
""" Returns a dictionary of power_name: [orders] for each power
The orders represent the opening orders that would have been submitted by the player
"""
game = Game()
state_proto = extract_state_proto(game)
phase_history_proto = extract_phase_history_proto(game)
possible_orders_proto = extract_possible_orders_proto(game)
# Retrieving all orders
# Not using kwargs - Using default player_seed, noise, temperature, and dropout_rate.
power_orders = yield [
self.get_orders_with_proto(state_proto,
power_name,
phase_history_proto,
possible_orders_proto,
retry_on_failure=False)
for power_name in game.powers
]
return {
power_name: orders
for power_name, orders in zip(game.powers.keys(), power_orders)
}
def __init__(self, env, players, clusters=None):
""" Constructor
:param env: The wrapped env
:param players: A list of player instances (all 7 players if no clusters, otherwise 1 player per cluster)
:param clusters: Optional. Contains a list of clusters, where each cluster is a tuple with
- 1) the number of players inside it, 2) a boolean that indicates if the players need to be merged
e.g. [(1,False), (3, False), (3, False)] would create a 1 vs 3 vs 3 game (3 clusters), where
players[0] is assigned to the first cluster,
players[1] is copied 3 times and assigned to the second cluster, and
players[2] is copied 3 times and assigned to the third cluster.
:type env: gym.core.Env
:type players: List[diplomacy_research.players.player.Player]
"""
super(RandomizePlayers, self).__init__(env)
game = self.game or Game()
self._powers = [power_name for power_name in game.powers]
self._clusters = clusters if clusters is not None else [
(1, False)
] * len(self._powers)
# Making sure we have the right number of players
if len(players) < len(self._clusters):
LOGGER.error(
'The nb of players (%d) must be greater or equal to %d.',
len(players), len(self._clusters))
raise ValueError()
# Generating the list of players based on the clusters definition
self._players = []
for cluster_ix, cluster_def in enumerate(self._clusters):
nb_players_in_cluster = cluster_def[0]
self._players += [players[cluster_ix]] * nb_players_in_cluster
# Generating an ordering to shuffle players
self._ordering = list(range(len(self._players)))
def state_dict_to_game_and_power(state_dict, country_id, max_phases=None):
""" Converts a game state from the dictionary format to an actual diplomacy.Game object with the related power.
:param state_dict: The game state in dictionary format from webdiplomacy.net
:param country_id: The country id we want to convert.
:param max_phases: Optional. If set, improve speed by only keeping the last 'x' phases to regenerate the game.
:return: A tuple of
1) None, None - on error or if the conversion is not possible, or game is invalid / not-started / done
2) game, power_name - on successful conversion
"""
if state_dict is None:
return None, None
req_fields = ('gameID', 'variantID', 'turn', 'phase', 'gameOver', 'phases', 'standoffs', 'occupiedFrom')
if [1 for field in req_fields if field not in state_dict]:
LOGGER.error('The required fields for state dict are %s. Cannot translate %s', req_fields, state_dict)
return None, None
# Extracting information
game_id = str(state_dict['gameID'])
map_id = int(state_dict['variantID'])
standoffs = state_dict['standoffs']
occupied_from = state_dict['occupiedFrom']
# Parsing all phases
state_dict_phases = state_dict.get('phases', [])
if max_phases is not None and isinstance(max_phases, int):
state_dict_phases = state_dict_phases[-1 * max_phases:]
all_phases = [process_phase_dict(phase_dict, map_id=map_id) for phase_dict in state_dict_phases]
# Building game - Replaying the last phases
game = Game(game_id=game_id, map_name=CACHE['ix_to_map'][map_id])
for phase_to_replay in all_phases[:-1]:
game.set_current_phase(phase_to_replay['name'])
# Units
game.clear_units()
for power_name, power_units in phase_to_replay['units'].items():
if power_name == 'GLOBAL':
continue
game.set_units(power_name, power_units)
# Centers
game.clear_centers()
for power_name, power_centers in phase_to_replay['centers'].items():
if power_name == 'GLOBAL':
continue
game.set_centers(power_name, power_centers)
# Orders
for power_name, power_orders in phase_to_replay['orders'].items():
if power_name == 'GLOBAL':
continue
game.set_orders(power_name, power_orders)
# Processing
game.process()
# Setting the current phase
current_phase = all_phases[-1]
game.set_current_phase(current_phase['name'])
# Units
game.clear_units()
for power_name, power_units in current_phase['units'].items():
if power_name == 'GLOBAL':
continue
game.set_units(power_name, power_units)
# Centers
game.clear_centers()
for power_name, power_centers in current_phase['centers'].items():
if power_name == 'GLOBAL':
continue
game.set_centers(power_name, power_centers)
# Setting retreat locs
if current_phase['name'][-1] == 'R':
invalid_retreat_locs = set()
attack_source = {}
# Loc is occupied
for power in game.powers.values():
for unit in power.units:
invalid_retreat_locs.add(unit[2:5])
# Loc was in standoff
if standoffs:
for loc_dict in standoffs:
_, loc = center_dict_to_str(loc_dict, map_id=map_id)
invalid_retreat_locs.add(loc[:3])
# Loc was attacked from
if occupied_from:
for loc_id, occupied_from_id in occupied_from.items():
loc_name = CACHE[map_id]['ix_to_loc'][int(loc_id)][:3]
from_loc_name = CACHE[map_id]['ix_to_loc'][int(occupied_from_id)][:3]
attack_source[loc_name] = from_loc_name
# Removing invalid retreat locs
for power in game.powers.values():
for retreat_unit in power.retreats:
power.retreats[retreat_unit] = [loc for loc in power.retreats[retreat_unit]
if loc[:3] not in invalid_retreat_locs
and loc[:3] != attack_source.get(retreat_unit[2:5], '')]
# Returning
power_name = CACHE[map_id]['ix_to_power'][country_id]
return game, power_name
def render_saved_game(saved_game, output_dir, prefix=''):
""" Renders a specific saved game
:param saved_game: The saved game to render
:param output_dir: The output directory where to save the rendering
:param prefix: An optional prefix to add before the game id
"""
if prefix:
output_dir = os.path.join(output_dir, prefix + '_' + saved_game['id'])
else:
output_dir = os.path.join(output_dir, saved_game['id'])
nb_phases = len(saved_game['phases'])
svg_count = 0
# Checking if already generated
# Otherwise, regenerating completely
if os.path.exists(output_dir):
nb_svg = len([
os.path.join(output_dir, file) for file in os.listdir(output_dir)
if file[-4:] == '.svg'
])
if nb_svg == 2 * nb_phases:
print('Rendered {} (Skipped)'.format(saved_game['id']))
return
shutil.rmtree(output_dir, ignore_errors=True)
os.makedirs(output_dir, exist_ok=True)
# Creating a Game to replay all orders, and a new Game object per phase to validate
entire_game = Game()
if saved_game['phases']:
entire_game.set_state(saved_game['phases'][0]['state'])
# Rendering
for phase in saved_game['phases']:
phase_game = Game()
# Setting state
state = phase['state']
phase_game.set_state(state)
entire_game.note = phase_game.note
# Setting orders
phase_game.clear_orders()
orders = phase['orders']
for power_name in orders:
phase_game.set_orders(power_name, orders[power_name])
entire_game.set_orders(power_name, orders[power_name])
# Validating that we are at the same place
for power_name in orders:
assert sorted(phase_game.get_units(power_name)) == sorted(
entire_game.get_units(power_name))
assert sorted(phase_game.get_centers(power_name)) == sorted(
entire_game.get_centers(power_name))
# Rendering with and without orders
with open(os.path.join(output_dir, '%03d%s' % (svg_count, '.svg')),
'w') as file:
file.write(entire_game.render(incl_orders=False))
svg_count += 1
with open(os.path.join(output_dir, '%03d%s' % (svg_count, '.svg')),
'w') as file:
file.write(entire_game.render(incl_orders=True))
# Processing (for entire game)
svg_count += 1
entire_game.process()
print('Rendered {}'.format(saved_game['id']))
def generate_trajectory(self):
game = Game()
powers = list(game.powers)
np.random.shuffle(powers)
power1 = powers[0]
powers_others = powers[1:]
action_probs = []
orders = []
values = []
supply_centers = [{power1: game.get_centers(power1)}]
while not game.is_game_done:
order, action_prob = self.actor.get_orders(game, [power1])
orders_others = {
power_name: self.actor.get_orders(game, [power_name])
for power_name in powers_others
}
board = tf.convert_to_tensor(
state_space.dict_to_flatten_board_state(
game.get_state(), game.map),
dtype=tf.float32)
board = tf.reshape(board, (1, 81 * 35))
print("TEST")
state_value = self.critic.call(board)
# Indexing because get_orders can return a list of lists orders for multiple powers
game.set_orders(power1, order[0])
for power_name, power_orders in orders_others.items():
orders_list, probs = power_orders
print(orders_list)
game.set_orders(power_name, orders_list[0])
game.process()
# Collect data
supply_centers.append({power1: game.get_centers(power1)})
action_probs.append(action_prob)
orders.append(order)
values.append(state_value)
# local_rewards.append(reward_class.get_local_reward(power1))
# global_rewards.append(0 if not game.is_game_done else reward_class.get_terminal_reward(power1))
rewards = get_average_reward([supply_centers])
returns = get_returns([
supply_centers
]) # put in list to match shape of [bs, game_length, dict}
return action_probs, returns, values, rewards
