def compute_action(self, state):
# convert tensor for rl
builder = PokerTreeBuilder()
state_tensor = builder.statenode_to_tensor(state)
# !!!! the return action is a longTensor[[]]
# action_id = (self.table_sl.select_action(state) if random.random() > arguments.eta \
# else self.dqn_optim.select_action(state_tensor))[0][0]
# action_id = self.table_sl.select_action(state)[0][0]
action_id = self.net_sl[state.node.current_player].select_action(
state_tensor).item()
# print('_____________')
# print(action_id)
# print('_____________')
# action['action: ,'raise_amount': ]
action = {}
#fold
if action_id == 0:
action['action'] = constants.acpc_actions.fold
# call
elif action_id == 1 or action_id >= game_settings.actions_count:
action['action'] = constants.acpc_actions.ccall
#raise
elif action_id > 1:
# get possible to determine the raising size
bet_sizding = BetSizing(arguments.Tensor(arguments.bet_sizing))
possible_bets = bet_sizding.get_possible_bets(state.node)
if possible_bets.dim() != 0:
possible_bet = possible_bets[:, state.node.current_player]
else:
action['action'] = constants.acpc_actions.ccall
return action
raise_action_id = action_id - 2 # to override fold and call action
# node possible bet in this state so call
action['action'] = constants.acpc_actions.rraise
if (len(possible_bet) <= raise_action_id):
action['raise_amount'] = possible_bet[len(possible_bet) -
1].item()
else:
action['raise_amount'] = possible_bet[raise_action_id].item(
) # to override fold and call action
else:
assert (False) #invaild actions
return action
params['root_node']['board'] = card_to_string.string_to_board('')
params['root_node']['street'] = 0
params['root_node']['current_player'] = constants.players.P1
params['root_node']['bets'] = arguments.Tensor([100, 100])
params['limit_to_street'] = False
tree = builder.build_tree(params)
table_sl = torch.load('/home/mjb/Nutstore/deepStack/Data/Model/Iter:' +
str(model_num) + '.sl')
dqn = DQN()
if torch.cuda.is_available():
dqn.cuda()
dqn.load_state_dict(
torch.load('/home/mjb/Nutstore/deepStack/Data/Model/Iter:' +
str(model_num) + '.rl'))
#dfs_fill_table(tree, table_sl, dqn, builder)
acc_list = []
acc_node = {}
builder.acc_node(tree, acc_node, acc_list)
state = GameState()
state.private = [arguments.Tensor([0]), arguments.Tensor([2])]
state.node = acc_node['858']
print(dqn(Variable(builder.statenode_to_tensor(state))))
#visualiser = TreeVisualiser()
#visualiser.graphviz(tree,"table_sl:" + str(model_num))
#!/usr/bin/env python2 # -*- coding: utf-8 -*- """ Created on Tue Aug 22 01:30:35 2017 @author: mjb """ import Settings.game_settings as game_settings import Game.card_to_string as card_to_string import Settings.arguments as arguments from Tree.tree_builder import PokerTreeBuilder from Tree.tree_visualiser import TreeVisualiser import Settings.constants as constants from nn.state import GameState from nn.env import Env builder = PokerTreeBuilder() env = Env() env.reset() tensor = builder.statenode_to_tensor(env.state) print(tensor.size(0))
