env.add_player(3, stack=20000) # add a player to seat 2 with 2000 "chips"
card_dictionary = get_card_dict()
for epoch in range(num_versions):
rl_dir = checkpoint_dir + '/rl_checkpoints/checkpoint_' + str(epoch +
1) + '000.pt'
sl_dir = checkpoint_dir + '/sl_checkpoints/checkpoint_' + str(epoch +
1) + '000.pt'
# policy_net = DQN(num_player=num_player,big=False).to(device)
policy_net = DQN_limit(num_player=num_player,
big=big,
num_action=3,
hidden_units=hid).to(device)
# checkpoint = torch.load('../holdem_result/rl_checkpoints/checkpoint1999999_8235201.000.pt')
checkpoint = torch.load(rl_dir, map_location='cpu')
policy_net.load_state_dict(checkpoint['model'])
policy_net.eval()
sl_net = MLP_limit(num_player=num_player,
big=big,
num_action=3,
hidden_units=hid).to(device)
# sl_net = MLP(num_player=num_player, big=False).to(device)
checkpoint = torch.load(sl_dir, map_location='cpu')
# checkpoint = torch.load('../holdem_result/sl_checkpoints/checkpoint1999999_8235201.000.pt')
sl_net.load_state_dict(checkpoint['model'])
sl_net.eval()
results_1 = []
results_2 = []
for expriment in range(3):
# networks
# networks
policy_net = DQN_limit(num_player=num_player,
big=big,
res_net=use_res_net,
num_layer=num_layer,
num_action=3,
hidden_units=num_hid).to(device)
target_net = DQN_limit(num_player=num_player,
big=big,
res_net=use_res_net,
num_layer=num_layer,
num_action=3,
hidden_units=num_hid).to(device)
target_net.load_state_dict(policy_net.state_dict())
target_net.eval()
sl_net = MLP_limit(num_player=num_player,
big=big,
res_net=use_res_net,
num_layer=num_layer,
num_action=3,
hidden_units=num_hid).to(device)
# optimizer
rl_optimizer = optim.SGD(policy_net.parameters(),
lr=lr_RL,
weight_decay=weight_decay)
sl_optimizer = optim.Adam(sl_net.parameters(),
lr=lr_SL,
weight_decay=weight_decay)