How to use the rlcard.agents.nfsp_agent.NFSPAgent function in rlcard

To help you get started, we’ve selected a few rlcard examples, based on popular ways it is used in public projects.

Secure your code as it's written. Use Snyk Code to scan source code in minutes - no build needed - and fix issues immediately.

github datamllab / rlcard / tests / agents / test_nfsp.py View on Github external
def test_init(self):

        sess = tf.InteractiveSession()
        tf.Variable(0, name='global_step', trainable=False)

        agent = NFSPAgent(sess=sess,
                         scope='nfsp',
                         action_num=10,
                         state_shape=[10],
                         hidden_layers_sizes=[10,10],
                         q_mlp_layers=[10,10])

        self.assertEqual(agent._action_num, 10)

        sess.close()
        tf.reset_default_graph()
github datamllab / rlcard / tests / agents / test_nfsp.py View on Github external
def test_evaluate_with(self):
        # Test average policy and value error here
        sess = tf.InteractiveSession()
        tf.Variable(0, name='global_step', trainable=False)

        agent = NFSPAgent(sess=sess,
                         scope='nfsp',
                         action_num=2,
                         state_shape=[2],
                         hidden_layers_sizes=[10,10],
                         q_mlp_layers=[10,10],
                         evaluate_with='average_policy')
        sess.run(tf.global_variables_initializer())
        predicted_action = agent.eval_step({'obs': np.random.random_sample((2,)), 'legal_actions': [0, 1]})
        self.assertGreaterEqual(predicted_action, 0)
        self.assertLessEqual(predicted_action, 1)

        sess.close()
        tf.reset_default_graph()

        sess = tf.InteractiveSession()
        tf.Variable(0, name='global_step', trainable=False)
github datamllab / rlcard / tests / agents / test_nfsp.py View on Github external
def test_train(self):

        norm_step = 100
        memory_init_size = 20
        step_num = 1000

        sess = tf.InteractiveSession()
        tf.Variable(0, name='global_step', trainable=False)
        agent = NFSPAgent(sess=sess,
                         scope='nfsp',
                         action_num=2,
                         state_shape=[2],
                         hidden_layers_sizes=[10,10],
                         reservoir_buffer_capacity=50,
                         batch_size=4,
                         min_buffer_size_to_learn=memory_init_size,
                         q_replay_memory_size=50,
                         q_replay_memory_init_size=memory_init_size,
                         q_batch_size=4,
                         q_norm_step=norm_step,
                         q_mlp_layers=[10,10])
        sess.run(tf.global_variables_initializer())

        predicted_action = agent.eval_step({'obs': np.random.random_sample((2,)), 'legal_actions': [0, 1]})
        self.assertGreaterEqual(predicted_action, 0)
github datamllab / rlcard / rlcard / models / pretrained_models.py View on Github external
def __init__(self):
        ''' Load pretrained model
        '''

        self.graph = tf.Graph()
        self.sess = tf.Session(graph=self.graph)

        env = rlcard.make('leduc-holdem')
        with self.graph.as_default():
            self.agents = []
            for i in range(env.player_num):
                agent = NFSPAgent(self.sess,
                                  scope='nfsp' + str(i),
                                  action_num=env.action_num,
                                  state_shape=env.state_shape,
                                  hidden_layers_sizes=[128,128],
                                  q_norm_step=1000,
                                  q_mlp_layers=[128,128])
                self.agents.append(agent)
            normalize(env, self.agents, 1000)
            self.sess.run(tf.global_variables_initializer())

        check_point_path = os.path.join(ROOT_PATH, 'leduc_holdem_nfsp')
        with self.sess.as_default():
            with self.graph.as_default():
                saver = tf.train.Saver(tf.model_variables())
                saver.restore(self.sess, tf.train.latest_checkpoint(check_point_path))
github datamllab / rlcard / examples / uno_nfsp.py View on Github external
# The paths for saving the logs and learning curves
root_path = './experiments/uno_nfsp_result/'
log_path = root_path + 'log.txt'
csv_path = root_path + 'performance.csv'
figure_path = root_path + 'figures/'

# Set a global seed
set_global_seed(0)

with tf.Session() as sess:
    # Set agents
    global_step = tf.Variable(0, name='global_step', trainable=False)
    agents = []
    for i in range(env.player_num):
        agent = NFSPAgent(sess,
                          scope='nfsp' + str(i),
                          action_num=env.action_num,
                          state_shape=env.state_shape,
                          hidden_layers_sizes=[512,1024,2048,1024,512],
                          anticipatory_param=0.5,
                          batch_size=256,
                          rl_learning_rate=0.00005,
                          sl_learning_rate=0.00001,
                          min_buffer_size_to_learn=memory_init_size,
                          q_replay_memory_size=int(1e5),
                          q_replay_memory_init_size=memory_init_size,
                          q_norm_step=norm_step,
                          q_batch_size=256,
                          q_mlp_layers=[512,1024,2048,1024,512])
        agents.append(agent)
github datamllab / rlcard / examples / nolimit_holdem_nfsp.py View on Github external
episode_num = 10000000

# Set the the number of steps for collecting normalization statistics
# and intial memory size
memory_init_size = 1000
norm_step = 1000

# Set a global seed
set_global_seed(0)

with tf.Session() as sess:
    # Set agents
    global_step = tf.Variable(0, name='global_step', trainable=False)
    agents = []
    for i in range(env.player_num):
        agent = NFSPAgent(sess,
                          scope='nfsp' + str(i),
                          action_num=env.action_num,
                          state_shape=[52],
                          hidden_layers_sizes=[512,512],
                          min_buffer_size_to_learn=memory_init_size,
                          q_replay_memory_init_size=memory_init_size,
                          q_norm_step=norm_step,
                          q_mlp_layers=[512,512])
        agents.append(agent)

    sess.run(tf.global_variables_initializer())

    random_agent = RandomAgent(action_num=eval_env.action_num)

    env.set_agents(agents)
    eval_env.set_agents([agents[0], random_agent])
github datamllab / rlcard / examples / limit_holdem_nfsp.py View on Github external
# The paths for saving the logs and learning curves
root_path = './experiments/limit_holdem_nfsp_result/'
log_path = root_path + 'log.txt'
csv_path = root_path + 'performance.csv'
figure_path = root_path + 'figures/'

# Set a global seed
set_global_seed(0)

with tf.Session() as sess:
    # Set agents
    global_step = tf.Variable(0, name='global_step', trainable=False)
    agents = []
    for i in range(env.player_num):
        agent = NFSPAgent(sess,
                          scope='nfsp' + str(i),
                          action_num=env.action_num,
                          state_shape=env.state_shape,
                          hidden_layers_sizes=[512,512],
                          anticipatory_param=0.1,
                          min_buffer_size_to_learn=memory_init_size,
                          q_replay_memory_init_size=memory_init_size,
                          q_norm_step=norm_step,
                          q_mlp_layers=[512,512])
        agents.append(agent)

    sess.run(tf.global_variables_initializer())

    random_agent = RandomAgent(action_num=eval_env.action_num)

    env.set_agents(agents)
github datamllab / rlcard / examples / doudizhu_nfsp.py View on Github external
episode_num = 10000000

# Set the the number of steps for collecting normalization statistics
# and intial memory size
memory_init_size = 1000
norm_step = 1000

# Set a global seed
set_global_seed(0)

with tf.Session() as sess:
    # Set agents
    global_step = tf.Variable(0, name='global_step', trainable=False)
    agents = []
    for i in range(env.player_num):
        agent = NFSPAgent(sess,
                          scope='nfsp' + str(i),
                          action_num=env.action_num,
                          state_shape=[6, 5, 15],
                          hidden_layers_sizes=[512,1024,2048,1024,512],
                          anticipatory_param=0.5,
                          batch_size=256,
                          rl_learning_rate=0.00005,
                          sl_learning_rate=0.00001,
                          min_buffer_size_to_learn=memory_init_size,
                          q_replay_memory_size=int(1e5),
                          q_replay_memory_init_size=memory_init_size,
                          q_norm_step=norm_step,
                          q_batch_size=256,
                          q_mlp_layers=[512,1024,2048,1024,512])
        agents.append(agent)
github datamllab / rlcard / examples / mahjong_nfsp.py View on Github external
# The paths for saving the logs and learning curves
root_path = './experiments/mahjong_nfsp_result/'
log_path = root_path + 'log.txt'
csv_path = root_path + 'performance.csv'
figure_path = root_path + 'figures/'

# Set a global seed
set_global_seed(0)

with tf.Session() as sess:
    # Set agents
    global_step = tf.Variable(0, name='global_step', trainable=False)
    agents = []
    for i in range(env.player_num):
        agent = NFSPAgent(sess,
                          scope='nfsp' + str(i),
                          action_num=env.action_num,
                          state_shape=env.state_shape,
                          hidden_layers_sizes=[512,1024,2048,1024,512],
                          anticipatory_param=0.5,
                          batch_size=256,
                          rl_learning_rate=0.00005,
                          sl_learning_rate=0.00001,
                          min_buffer_size_to_learn=memory_init_size,
                          q_replay_memory_size=int(1e5),
                          q_replay_memory_init_size=memory_init_size,
                          q_norm_step=norm_step,
                          q_batch_size=256,
                          q_mlp_layers=[512,1024,2048,1024,512])
        agents.append(agent)
github datamllab / rlcard / examples / leduc_holdem_nfsp.py View on Github external
episode_num = 10000000

# Set the the number of steps for collecting normalization statistics
# and intial memory size
memory_init_size = 1000
norm_step = 1000

# Set a global seed
set_global_seed(0)

with tf.Session() as sess:
    # Set agents
    global_step = tf.Variable(0, name='global_step', trainable=False)
    agents = []
    for i in range(env.player_num):
        agent = NFSPAgent(sess,
                          scope='nfsp' + str(i),
                          action_num=env.action_num,
                          state_shape=[6],
                          hidden_layers_sizes=[128,128],
                          min_buffer_size_to_learn=memory_init_size,
                          q_replay_memory_init_size=memory_init_size,
                          q_norm_step=norm_step,
                          q_mlp_layers=[128,128])
        agents.append(agent)

    sess.run(tf.global_variables_initializer())

    random_agent = RandomAgent(action_num=eval_env.action_num)

    env.set_agents(agents)
    eval_env.set_agents([agents[0], random_agent])