def get_observation_size(game_config):
""" Returns the len of the vectorized observation """
num_players = game_config['num_total_players'] # number of players ingame
num_colors = game_config['colors']
num_ranks = game_config['ranks']
hand_size = game_config['hand_size']
max_information_tokens = game_config['life_tokens']
max_life_tokens = game_config['info_tokens']
max_moves = game_config['max_moves']
variant = game_config['variant']
env = json_to_pyhanabi.create_env_mock(
num_players=num_players,
num_colors=num_colors,
num_ranks=num_ranks,
hand_size=hand_size,
max_information_tokens=max_information_tokens,
max_life_tokens=max_life_tokens,
max_moves=max_moves,
variant=variant)
vec = vectorizer.ObservationVectorizer(env)
legal_moves_vectorizer = vectorizer.LegalMovesVectorizer(env)
return vec.total_state_length
def __init__(self, game_config):
"""
# ################################################ #
# -------------------- CONFIG -------------------- #
# ################################################ #
"""
self.agent_name = game_config[
'username'] # used to identify absolute position on table
self.num_players = game_config[
'num_total_players'] # number of players ingame
self.max_life_tokens = game_config['life_tokens']
self.max_info_tokens = game_config['info_tokens']
self.max_deck_size = game_config['deck_size']
self.deck_size = self.max_deck_size
self.life_tokens = self.max_life_tokens
self.information_tokens = self.max_info_tokens
self.players = None # list of names of players currently ingame
self.player_position = None # agents absolute position at table
self.agents_turn = False # flag that is True whenever its our turn
self.hand_size = 4 if self.num_players > 3 else 5 # deal 5 cards when playing with 2 or 3 ppl
"""
# ################################################ #
# ------- Observed Cards and Card knowledge ------ #
# ################################################ #
"""
""" New cards are prepended, that means agent 1s inital draw looks like [4,3,2,1] """
# list of all players hands as _seen_ by calling agent [excluding clues]
self.observed_hands = list(
) # is refreshed in self.update() on each notify message
# list of clues given
self.clues = list(
) # is refreshed in self.update() on each notify message
# unfortunately, server references clue cards not by index but by an id between 0 and deck size,
# so we need to store card_numbers to map the card ids to indices
self.card_numbers = list()
"""
# ################################################ #
# ----------------- GAME STATS ------------------- #
# ################################################ #
"""
# is refreshed in self.update() on each notify message
self.fireworks = {'R': 0, 'Y': 0, 'G': 0, 'W': 0, 'B': 0}
# list of discarded cards as returned by self.card(suit, rank)
self.discard_pile = list()
# actually not contained in the returned dict of the
# rl_env.HanabiEnvobservation._extract_from_dict()-method, but we need a history so we add this here.
# Similarly, it could be added by appending obs_dict['last_moves'] = observation.last_moves() in said method.
self.last_moves = list()
self.variant = game_config['variant']
self.num_colors = game_config['colors']
self.num_ranks = game_config['ranks']
self.max_moves = game_config['max_moves']
self.order = 0 # number of the next card on the deck, incremented when card is drawn/dealt
"""
# ################################################ #
# -------------- USE PYHANABI MOCKS -------------- #
# ################################################ #
"""
self.env = json_to_pyhanabi.create_env_mock(
num_players=self.num_players,
num_colors=self.num_colors,
num_ranks=self.num_ranks,
hand_size=self.hand_size,
max_information_tokens=self.max_info_tokens,
max_life_tokens=self.max_life_tokens,
max_moves=self.max_moves,
variant=self.variant)
self.caller_is_admin = False # flag is used to determine one RL agent that keeps track of human players
# environment observations in order to be able to keep the vectorized observations synchronized
if self.agent_name[-2:] == '00':
self.caller_is_admin = True # admin is the first instance of the client class
self.idx_human_player = -1 # used to determine whenever human is target of a card hint and thus when
# hints will be out of sync with vectorizer environment state
self.vectorizer_is_synced = False # if human player got card hints, the other vectorizer instances must know
self.vectorizer = vectorizer.ObservationVectorizer(self.env)
self.legal_moves_vectorizer = vectorizer.LegalMovesVectorizer(self.env)