update a3c_cartpole.py to tf2.x (#8662)

* update a3c_cartpole.py to tf2.x

* update a3c_cartpole.py to tf2.x

research/a3c_blogpost/README.md

@@ -3,4 +3,4 @@ In order to run this code, you will need the following prerequisites:
 
 * [OpenAI Gym](https://github.com/openai/gym) - `pip install gym`
 * [pyglet](https://bitbucket.org/pyglet/pyglet/wiki/Home) - `pip install pyglet` 
-* [TensorFlow](https://www.tensorflow.org/install/) - `pip install tensorflow==v1.14.0`
+* [TensorFlow](https://www.tensorflow.org/install/) - `pip install tensorflow==2.2.0`

research/a3c_blogpost/a3c_cartpole.py

@@ -14,8 +14,6 @@ import tensorflow as tf
 from tensorflow.python import keras
 from tensorflow.python.keras import layers
 
-tf.enable_eager_execution()
-
 parser = argparse.ArgumentParser(description='Run A3C algorithm on the game '
                                              'Cartpole.')
 parser.add_argument('--algorithm', default='a3c', type=str,
@@ -135,7 +133,7 @@ class MasterAgent():
     env = gym.make(self.game_name)
     self.state_size = env.observation_space.shape[0]
     self.action_size = env.action_space.n
-    self.opt = tf.train.AdamOptimizer(args.lr, use_locking=True)
+    self.opt = tf.compat.v1.train.AdamOptimizer(args.lr, use_locking=True)
     print(self.state_size, self.action_size)
 
     self.global_model = ActorCriticModel(self.state_size, self.action_size)  # global network
@@ -348,7 +346,7 @@ class Worker(threading.Thread):
 
     # Calculate our policy loss
     policy = tf.nn.softmax(logits)
-    entropy = tf.nn.softmax_cross_entropy_with_logits_v2(labels=policy, logits=logits)
+    entropy = tf.nn.softmax_cross_entropy_with_logits(labels=policy, logits=logits)
 
     policy_loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=memory.actions,
                                                                  logits=logits)