A3C_distributed_tf.py

"""
Asynchronous Advantage Actor Critic (A3C) with discrete action space, Reinforcement Learning.

The Cartpole example using distributed tensorflow + multiprocessing.

View more on my tutorial page: https://morvanzhou.github.io/

"""

import multiprocessing as mp
import tensorflow as tf
import numpy as np
import gym, time
import matplotlib.pyplot as plt


UPDATE_GLOBAL_ITER = 10
GAMMA = 0.9
ENTROPY_BETA = 0.001
LR_A = 0.001    # learning rate for actor
LR_C = 0.001    # learning rate for critic

env = gym.make('CartPole-v0')
N_S = env.observation_space.shape[0]
N_A = env.action_space.n


class ACNet(object):
    sess = None

    def __init__(self, scope, opt_a=None, opt_c=None, global_net=None):
        if scope == 'global_net':  # get global network
            with tf.variable_scope(scope):
                self.s = tf.placeholder(tf.float32, [None, N_S], 'S')
                self.a_params, self.c_params = self._build_net(scope)[-2:]
        else:
            with tf.variable_scope(scope):
                self.s = tf.placeholder(tf.float32, [None, N_S], 'S')
                self.a_his = tf.placeholder(tf.int32, [None, ], 'A')
                self.v_target = tf.placeholder(tf.float32, [None, 1], 'Vtarget')

                self.a_prob, self.v, self.a_params, self.c_params = self._build_net(scope)

                td = tf.subtract(self.v_target, self.v, name='TD_error')
                with tf.name_scope('c_loss'):
                    self.c_loss = tf.reduce_mean(tf.square(td))

                with tf.name_scope('a_loss'):
                    log_prob = tf.reduce_sum(
                        tf.log(self.a_prob) * tf.one_hot(self.a_his, N_A, dtype=tf.float32),
                        axis=1, keep_dims=True)
                    exp_v = log_prob * tf.stop_gradient(td)
                    entropy = -tf.reduce_sum(self.a_prob * tf.log(self.a_prob + 1e-5),
                                             axis=1, keep_dims=True)  # encourage exploration
                    self.exp_v = ENTROPY_BETA * entropy + exp_v
                    self.a_loss = tf.reduce_mean(-self.exp_v)

                with tf.name_scope('local_grad'):
                    self.a_grads = tf.gradients(self.a_loss, self.a_params)
                    self.c_grads = tf.gradients(self.c_loss, self.c_params)

            self.global_step = tf.train.get_or_create_global_step()
            with tf.name_scope('sync'):
                with tf.name_scope('pull'):
                    self.pull_a_params_op = [l_p.assign(g_p) for l_p, g_p in zip(self.a_params, global_net.a_params)]
                    self.pull_c_params_op = [l_p.assign(g_p) for l_p, g_p in zip(self.c_params, global_net.c_params)]
                with tf.name_scope('push'):
                    self.update_a_op = opt_a.apply_gradients(zip(self.a_grads, global_net.a_params), global_step=self.global_step)
                    self.update_c_op = opt_c.apply_gradients(zip(self.c_grads, global_net.c_params))

    def _build_net(self, scope):
        w_init = tf.random_normal_initializer(0., .1)
        with tf.variable_scope('actor'):
            l_a = tf.layers.dense(self.s, 200, tf.nn.relu6, kernel_initializer=w_init, name='la')
            a_prob = tf.layers.dense(l_a, N_A, tf.nn.softmax, kernel_initializer=w_init, name='ap')
        with tf.variable_scope('critic'):
            l_c = tf.layers.dense(self.s, 100, tf.nn.relu6, kernel_initializer=w_init, name='lc')
            v = tf.layers.dense(l_c, 1, kernel_initializer=w_init, name='v')  # state value
        a_params = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope=scope + '/actor')
        c_params = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope=scope + '/critic')
        return a_prob, v, a_params, c_params

    def choose_action(self, s):  # run by a local
        prob_weights = self.sess.run(self.a_prob, feed_dict={self.s: s[np.newaxis, :]})
        action = np.random.choice(range(prob_weights.shape[1]),
                                  p=prob_weights.ravel())  # select action w.r.t the actions prob
        return action

    def update_global(self, feed_dict):  # run by a local
        self.sess.run([self.update_a_op, self.update_c_op], feed_dict)  # local grads applies to global net

    def pull_global(self):  # run by a local
        self.sess.run([self.pull_a_params_op, self.pull_c_params_op])


def work(job_name, task_index, global_ep, lock, r_queue, global_running_r):
    # set work's ip:port
    cluster = tf.train.ClusterSpec({
        "ps": ['localhost:2220', 'localhost:2221',],
        "worker": ['localhost:2222', 'localhost:2223', 'localhost:2224', 'localhost:2225',]
    })
    server = tf.train.Server(cluster, job_name=job_name, task_index=task_index)
    if job_name == 'ps':
        print('Start Parameter Sever: ', task_index)
        server.join()
    else:
        t1 = time.time()
        env = gym.make('CartPole-v0').unwrapped
        print('Start Worker: ', task_index)
        with tf.device(tf.train.replica_device_setter(
                worker_device="/job:worker/task:%d" % task_index,
                cluster=cluster)):
            opt_a = tf.train.RMSPropOptimizer(LR_A, name='opt_a')
            opt_c = tf.train.RMSPropOptimizer(LR_C, name='opt_c')
            global_net = ACNet('global_net')

        local_net = ACNet('local_ac%d' % task_index, opt_a, opt_c, global_net)
        # set training steps
        hooks = [tf.train.StopAtStepHook(last_step=100000)]
        with tf.train.MonitoredTrainingSession(master=server.target,
                                               is_chief=True,
                                               hooks=hooks,) as sess:
            print('Start Worker Session: ', task_index)
            local_net.sess = sess
            total_step = 1
            buffer_s, buffer_a, buffer_r = [], [], []
            while (not sess.should_stop()) and (global_ep.value < 1000):
                s = env.reset()
                ep_r = 0
                while True:
                    # if task_index:
                    #     env.render()
                    a = local_net.choose_action(s)
                    s_, r, done, info = env.step(a)
                    if done: r = -5.
                    ep_r += r
                    buffer_s.append(s)
                    buffer_a.append(a)
                    buffer_r.append(r)

                    if total_step % UPDATE_GLOBAL_ITER == 0 or done:  # update global and assign to local net
                        if done:
                            v_s_ = 0  # terminal
                        else:
                            v_s_ = sess.run(local_net.v, {local_net.s: s_[np.newaxis, :]})[0, 0]
                        buffer_v_target = []
                        for r in buffer_r[::-1]:  # reverse buffer r
                            v_s_ = r + GAMMA * v_s_
                            buffer_v_target.append(v_s_)
                        buffer_v_target.reverse()

                        buffer_s, buffer_a, buffer_v_target = np.vstack(buffer_s), np.array(buffer_a), np.vstack(
                            buffer_v_target)
                        feed_dict = {
                            local_net.s: buffer_s,
                            local_net.a_his: buffer_a,
                            local_net.v_target: buffer_v_target,
                        }
                        local_net.update_global(feed_dict)
                        buffer_s, buffer_a, buffer_r = [], [], []
                        local_net.pull_global()
                    s = s_
                    total_step += 1
                    if done:
                        if r_queue.empty():  # record running episode reward
                            global_running_r.value = ep_r
                        else:
                            global_running_r.value = .99 * global_running_r.value + 0.01 * ep_r
                        r_queue.put(global_running_r.value)

                        print(
                            "Task: %i" % task_index,
                            "| Ep: %i" % global_ep.value,
                            "| Ep_r: %i" % global_running_r.value,
                            "| Global_step: %i" % sess.run(local_net.global_step),
                        )
                        with lock:
                            global_ep.value += 1
                        break

        print('Worker Done: ', task_index, time.time()-t1)


if __name__ == "__main__":
    # use multiprocessing to create a local cluster with 2 parameter servers and 2 workers
    global_ep = mp.Value('i', 0)
    lock = mp.Lock()
    r_queue = mp.Queue()
    global_running_r = mp.Value('d', 0)

    jobs = [
        ('ps', 0), ('ps', 1),
        ('worker', 0), ('worker', 1), ('worker', 2), ('worker', 3)
    ]
    ps = [mp.Process(target=work, args=(j, i, global_ep, lock, r_queue, global_running_r), ) for j, i in jobs]
    [p.start() for p in ps]
    [p.join() for p in ps[2:]]

    ep_r = []
    while not r_queue.empty():
        ep_r.append(r_queue.get())
    plt.plot(np.arange(len(ep_r)), ep_r)
    plt.title('Distributed training')
    plt.xlabel('Step')
    plt.ylabel('Total moving reward')
    plt.show()