Thanks for tuning in to Google I/O. View all sessions on demandWatch on demand

Train a Deep Q Network with TF-Agents

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook

Introduction

This example shows how to train a DQN (Deep Q Networks) agent on the Cartpole environment using the TF-Agents library.

Cartpole environment

It will walk you through all the components in a Reinforcement Learning (RL) pipeline for training, evaluation and data collection.

To run this code live, click the 'Run in Google Colab' link above.

Setup

If you haven't installed the following dependencies, run:

sudo apt-get update
sudo apt-get install -y xvfb ffmpeg freeglut3-dev
pip install 'imageio==2.4.0'
pip install pyvirtualdisplay
pip install tf-agents[reverb]
pip install pyglet

from __future__ import absolute_import, division, print_function

import base64
import imageio
import IPython
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import PIL.Image
import pyvirtualdisplay
import reverb

import tensorflow as tf

from tf_agents.agents.dqn import dqn_agent
from tf_agents.drivers import py_driver
from tf_agents.environments import suite_gym
from tf_agents.environments import tf_py_environment
from tf_agents.eval import metric_utils
from tf_agents.metrics import tf_metrics
from tf_agents.networks import sequential
from tf_agents.policies import py_tf_eager_policy
from tf_agents.policies import random_tf_policy
from tf_agents.replay_buffers import reverb_replay_buffer
from tf_agents.replay_buffers import reverb_utils
from tf_agents.trajectories import trajectory
from tf_agents.specs import tensor_spec
from tf_agents.utils import common

2023-02-16 13:29:03.229568: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer.so.7'; dlerror: libnvinfer.so.7: cannot open shared object file: No such file or directory
2023-02-16 13:29:03.229668: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer_plugin.so.7'; dlerror: libnvinfer_plugin.so.7: cannot open shared object file: No such file or directory
2023-02-16 13:29:03.229677: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.

# Set up a virtual display for rendering OpenAI gym environments.
display = pyvirtualdisplay.Display(visible=0, size=(1400, 900)).start()

tf.version.VERSION

'2.11.0'

Hyperparameters

num_iterations = 20000 # @param {type:"integer"}

initial_collect_steps = 100  # @param {type:"integer"}
collect_steps_per_iteration =   1# @param {type:"integer"}
replay_buffer_max_length = 100000  # @param {type:"integer"}

batch_size = 64  # @param {type:"integer"}
learning_rate = 1e-3  # @param {type:"number"}
log_interval = 200  # @param {type:"integer"}

num_eval_episodes = 10  # @param {type:"integer"}
eval_interval = 1000  # @param {type:"integer"}

Environment

In Reinforcement Learning (RL), an environment represents the task or problem to be solved. Standard environments can be created in TF-Agents using tf_agents.environments suites. TF-Agents has suites for loading environments from sources such as the OpenAI Gym, Atari, and DM Control.

Load the CartPole environment from the OpenAI Gym suite. 

env_name = 'CartPole-v0'
env = suite_gym.load(env_name)

You can render this environment to see how it looks. A free-swinging pole is attached to a cart. The goal is to move the cart right or left in order to keep the pole pointing up.

env.reset()
PIL.Image.fromarray(env.render())

png

The environment.step method takes an action in the environment and returns a TimeStep tuple containing the next observation of the environment and the reward for the action.

The time_step_spec() method returns the specification for the TimeStep tuple. Its observation attribute shows the shape of observations, the data types, and the ranges of allowed values. The reward attribute shows the same details for the reward.

print('Observation Spec:')
print(env.time_step_spec().observation)

Observation Spec:
BoundedArraySpec(shape=(4,), dtype=dtype('float32'), name='observation', minimum=[-4.8000002e+00 -3.4028235e+38 -4.1887903e-01 -3.4028235e+38], maximum=[4.8000002e+00 3.4028235e+38 4.1887903e-01 3.4028235e+38])

print('Reward Spec:')
print(env.time_step_spec().reward)

Reward Spec:
ArraySpec(shape=(), dtype=dtype('float32'), name='reward')

The action_spec() method returns the shape, data types, and allowed values of valid actions.

print('Action Spec:')
print(env.action_spec())

Action Spec:
BoundedArraySpec(shape=(), dtype=dtype('int64'), name='action', minimum=0, maximum=1)

In the Cartpole environment:

  • observation is an array of 4 floats:
    • the position and velocity of the cart
    • the angular position and velocity of the pole
  • reward is a scalar float value
  • action is a scalar integer with only two possible values:
    • 0 — "move left"
    • 1 — "move right"
time_step = env.reset()
print('Time step:')
print(time_step)

action = np.array(1, dtype=np.int32)

next_time_step = env.step(action)
print('Next time step:')
print(next_time_step)

Time step:
TimeStep(
{'discount': array(1., dtype=float32),
 'observation': array([-0.00597946,  0.02228957, -0.04941824, -0.02423982], dtype=float32),
 'reward': array(0., dtype=float32),
 'step_type': array(0, dtype=int32)})
Next time step:
TimeStep(
{'discount': array(1., dtype=float32),
 'observation': array([-0.00553366,  0.21808411, -0.04990304, -0.33209613], dtype=float32),
 'reward': array(1., dtype=float32),
 'step_type': array(1, dtype=int32)})

Usually two environments are instantiated: one for training and one for evaluation. 

train_py_env = suite_gym.load(env_name)
eval_py_env = suite_gym.load(env_name)

The Cartpole environment, like most environments, is written in pure Python. This is converted to TensorFlow using the TFPyEnvironment wrapper.

The original environment's API uses Numpy arrays. The TFPyEnvironment converts these to Tensors to make it compatible with Tensorflow agents and policies.

train_env = tf_py_environment.TFPyEnvironment(train_py_env)
eval_env = tf_py_environment.TFPyEnvironment(eval_py_env)

Agent

The algorithm used to solve an RL problem is represented by an Agent. TF-Agents provides standard implementations of a variety of Agents, including:

The DQN agent can be used in any environment which has a discrete action space.

At the heart of a DQN Agent is a QNetwork, a neural network model that can learn to predict QValues (expected returns) for all actions, given an observation from the environment.

We will use tf_agents.networks. to create a QNetwork. The network will consist of a sequence of tf.keras.layers.Dense layers, where the final layer will have 1 output for each possible action.

fc_layer_params = (100, 50)
action_tensor_spec = tensor_spec.from_spec(env.action_spec())
num_actions = action_tensor_spec.maximum - action_tensor_spec.minimum + 1

# Define a helper function to create Dense layers configured with the right
# activation and kernel initializer.
def dense_layer(num_units):
  return tf.keras.layers.Dense(
      num_units,
      activation=tf.keras.activations.relu,
      kernel_initializer=tf.keras.initializers.VarianceScaling(
          scale=2.0, mode='fan_in', distribution='truncated_normal'))

# QNetwork consists of a sequence of Dense layers followed by a dense layer
# with `num_actions` units to generate one q_value per available action as
# its output.
dense_layers = [dense_layer(num_units) for num_units in fc_layer_params]
q_values_layer = tf.keras.layers.Dense(
    num_actions,
    activation=None,
    kernel_initializer=tf.keras.initializers.RandomUniform(
        minval=-0.03, maxval=0.03),
    bias_initializer=tf.keras.initializers.Constant(-0.2))
q_net = sequential.Sequential(dense_layers + [q_values_layer])

Now use tf_agents.agents.dqn.dqn_agent to instantiate a DqnAgent. In addition to the time_step_spec, action_spec and the QNetwork, the agent constructor also requires an optimizer (in this case, AdamOptimizer), a loss function, and an integer step counter.

optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate)

train_step_counter = tf.Variable(0)

agent = dqn_agent.DqnAgent(
    train_env.time_step_spec(),
    train_env.action_spec(),
    q_network=q_net,
    optimizer=optimizer,
    td_errors_loss_fn=common.element_wise_squared_loss,
    train_step_counter=train_step_counter)

agent.initialize()

Policies

A policy defines the way an agent acts in an environment. Typically, the goal of reinforcement learning is to train the underlying model until the policy produces the desired outcome.

In this tutorial:

  • The desired outcome is keeping the pole balanced upright over the cart.
  • The policy returns an action (left or right) for each time_step observation.

Agents contain two policies:

  • agent.policy — The main policy that is used for evaluation and deployment.
  • agent.collect_policy — A second policy that is used for data collection.
eval_policy = agent.policy
collect_policy = agent.collect_policy

Policies can be created independently of agents. For example, use tf_agents.policies.random_tf_policy to create a policy which will randomly select an action for each time_step.

random_policy = random_tf_policy.RandomTFPolicy(train_env.time_step_spec(),
                                                train_env.action_spec())

To get an action from a policy, call the policy.action(time_step) method. The time_step contains the observation from the environment. This method returns a PolicyStep, which is a named tuple with three components:

  • action — the action to be taken (in this case, 0 or 1)
  • state — used for stateful (that is, RNN-based) policies
  • info — auxiliary data, such as log probabilities of actions
example_environment = tf_py_environment.TFPyEnvironment(
    suite_gym.load('CartPole-v0'))

time_step = example_environment.reset()

random_policy.action(time_step)

PolicyStep(action=<tf.Tensor: shape=(1,), dtype=int64, numpy=array([0])>, state=(), info=())

Metrics and Evaluation

The most common metric used to evaluate a policy is the average return. The return is the sum of rewards obtained while running a policy in an environment for an episode. Several episodes are run, creating an average return.

The following function computes the average return of a policy, given the policy, environment, and a number of episodes.

def compute_avg_return(environment, policy, num_episodes=10):

  total_return = 0.0
  for _ in range(num_episodes):

    time_step = environment.reset()
    episode_return = 0.0

    while not time_step.is_last():
      action_step = policy.action(time_step)
      time_step = environment.step(action_step.action)
      episode_return += time_step.reward
    total_return += episode_return

  avg_return = total_return / num_episodes
  return avg_return.numpy()[0]


# See also the metrics module for standard implementations of different metrics.
# https://github.com/tensorflow/agents/tree/master/tf_agents/metrics

Running this computation on the random_policy shows a baseline performance in the environment.

compute_avg_return(eval_env, random_policy, num_eval_episodes)

22.2

Replay Buffer

In order to keep track of the data collected from the environment, we will use Reverb, an efficient, extensible, and easy-to-use replay system by Deepmind. It stores experience data when we collect trajectories and is consumed during training.

This replay buffer is constructed using specs describing the tensors that are to be stored, which can be obtained from the agent using agent.collect_data_spec.

table_name = 'uniform_table'
replay_buffer_signature = tensor_spec.from_spec(
      agent.collect_data_spec)
replay_buffer_signature = tensor_spec.add_outer_dim(
    replay_buffer_signature)

table = reverb.Table(
    table_name,
    max_size=replay_buffer_max_length,
    sampler=reverb.selectors.Uniform(),
    remover=reverb.selectors.Fifo(),
    rate_limiter=reverb.rate_limiters.MinSize(1),
    signature=replay_buffer_signature)

reverb_server = reverb.Server([table])

replay_buffer = reverb_replay_buffer.ReverbReplayBuffer(
    agent.collect_data_spec,
    table_name=table_name,
    sequence_length=2,
    local_server=reverb_server)

rb_observer = reverb_utils.ReverbAddTrajectoryObserver(
  replay_buffer.py_client,
  table_name,
  sequence_length=2)

[reverb/cc/platform/tfrecord_checkpointer.cc:162]  Initializing TFRecordCheckpointer in /tmpfs/tmp/tmpfm37nz7v.
[reverb/cc/platform/tfrecord_checkpointer.cc:567] Loading latest checkpoint from /tmpfs/tmp/tmpfm37nz7v
[reverb/cc/platform/default/server.cc:71] Started replay server on port 39871

For most agents, collect_data_spec is a named tuple called Trajectory, containing the specs for observations, actions, rewards, and other items.

agent.collect_data_spec

Trajectory(
{'action': BoundedTensorSpec(shape=(), dtype=tf.int64, name='action', minimum=array(0), maximum=array(1)),
 'discount': BoundedTensorSpec(shape=(), dtype=tf.float32, name='discount', minimum=array(0., dtype=float32), maximum=array(1., dtype=float32)),
 'next_step_type': TensorSpec(shape=(), dtype=tf.int32, name='step_type'),
 'observation': BoundedTensorSpec(shape=(4,), dtype=tf.float32, name='observation', minimum=array([-4.8000002e+00, -3.4028235e+38, -4.1887903e-01, -3.4028235e+38],
      dtype=float32), maximum=array([4.8000002e+00, 3.4028235e+38, 4.1887903e-01, 3.4028235e+38],
      dtype=float32)),
 'policy_info': (),
 'reward': TensorSpec(shape=(), dtype=tf.float32, name='reward'),
 'step_type': TensorSpec(shape=(), dtype=tf.int32, name='step_type')})

agent.collect_data_spec._fields

('step_type',
 'observation',
 'action',
 'policy_info',
 'next_step_type',
 'reward',
 'discount')

Data Collection

Now execute the random policy in the environment for a few steps, recording the data in the replay buffer.

Here we are using 'PyDriver' to run the experience collecting loop. You can learn more about TF Agents driver in our drivers tutorial.

py_driver.PyDriver(
    env,
    py_tf_eager_policy.PyTFEagerPolicy(
      random_policy, use_tf_function=True),
    [rb_observer],
    max_steps=initial_collect_steps).run(train_py_env.reset())

(TimeStep(
 {'discount': array(1., dtype=float32),
  'observation': array([ 0.03317351,  0.03446898, -0.01081271, -0.09491368], dtype=float32),
  'reward': array(1., dtype=float32),
  'step_type': array(1, dtype=int32)}),
 ())

The replay buffer is now a collection of Trajectories.

# For the curious:
# Uncomment to peel one of these off and inspect it.
# iter(replay_buffer.as_dataset()).next()

The agent needs access to the replay buffer. This is provided by creating an iterable tf.data.Dataset pipeline which will feed data to the agent.

Each row of the replay buffer only stores a single observation step. But since the DQN Agent needs both the current and next observation to compute the loss, the dataset pipeline will sample two adjacent rows for each item in the batch (num_steps=2).

This dataset is also optimized by running parallel calls and prefetching data.

# Dataset generates trajectories with shape [Bx2x...]
dataset = replay_buffer.as_dataset(
    num_parallel_calls=3,
    sample_batch_size=batch_size,
    num_steps=2).prefetch(3)

dataset

WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/tensorflow/python/autograph/pyct/static_analysis/liveness.py:83: Analyzer.lamba_check (from tensorflow.python.autograph.pyct.static_analysis.liveness) is deprecated and will be removed after 2023-09-23.
Instructions for updating:
Lambda fuctions will be no more assumed to be used in the statement where they are used, or at least in the same block. https://github.com/tensorflow/tensorflow/issues/56089
<PrefetchDataset element_spec=(Trajectory(
{'action': TensorSpec(shape=(64, 2), dtype=tf.int64, name=None),
 'discount': TensorSpec(shape=(64, 2), dtype=tf.float32, name=None),
 'next_step_type': TensorSpec(shape=(64, 2), dtype=tf.int32, name=None),
 'observation': TensorSpec(shape=(64, 2, 4), dtype=tf.float32, name=None),
 'policy_info': (),
 'reward': TensorSpec(shape=(64, 2), dtype=tf.float32, name=None),
 'step_type': TensorSpec(shape=(64, 2), dtype=tf.int32, name=None)}), SampleInfo(key=TensorSpec(shape=(64, 2), dtype=tf.uint64, name=None), probability=TensorSpec(shape=(64, 2), dtype=tf.float64, name=None), table_size=TensorSpec(shape=(64, 2), dtype=tf.int64, name=None), priority=TensorSpec(shape=(64, 2), dtype=tf.float64, name=None), times_sampled=TensorSpec(shape=(64, 2), dtype=tf.int32, name=None)))>

iterator = iter(dataset)
print(iterator)

<tensorflow.python.data.ops.iterator_ops.OwnedIterator object at 0x7fa86fb85f70>

# For the curious:
# Uncomment to see what the dataset iterator is feeding to the agent.
# Compare this representation of replay data 
# to the collection of individual trajectories shown earlier.

# iterator.next()

Training the agent

Two things must happen during the training loop:

  • collect data from the environment
  • use that data to train the agent's neural network(s)

This example also periodicially evaluates the policy and prints the current score.

The following will take ~5 minutes to run.

try:
  %%time
except:
  pass

# (Optional) Optimize by wrapping some of the code in a graph using TF function.
agent.train = common.function(agent.train)

# Reset the train step.
agent.train_step_counter.assign(0)

# Evaluate the agent's policy once before training.
avg_return = compute_avg_return(eval_env, agent.policy, num_eval_episodes)
returns = [avg_return]

# Reset the environment.
time_step = train_py_env.reset()

# Create a driver to collect experience.
collect_driver = py_driver.PyDriver(
    env,
    py_tf_eager_policy.PyTFEagerPolicy(
      agent.collect_policy, use_tf_function=True),
    [rb_observer],
    max_steps=collect_steps_per_iteration)

for _ in range(num_iterations):

  # Collect a few steps and save to the replay buffer.
  time_step, _ = collect_driver.run(time_step)

  # Sample a batch of data from the buffer and update the agent's network.
  experience, unused_info = next(iterator)
  train_loss = agent.train(experience).loss

  step = agent.train_step_counter.numpy()

  if step % log_interval == 0:
    print('step = {0}: loss = {1}'.format(step, train_loss))

  if step % eval_interval == 0:
    avg_return = compute_avg_return(eval_env, agent.policy, num_eval_episodes)
    print('step = {0}: Average Return = {1}'.format(step, avg_return))
    returns.append(avg_return)

WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/tensorflow/python/util/dispatch.py:1176: calling foldr_v2 (from tensorflow.python.ops.functional_ops) with back_prop=False is deprecated and will be removed in a future version.
Instructions for updating:
back_prop=False is deprecated. Consider using tf.stop_gradient instead.
Instead of:
results = tf.foldr(fn, elems, back_prop=False)
Use:
results = tf.nest.map_structure(tf.stop_gradient, tf.foldr(fn, elems))
[reverb/cc/client.cc:165] Sampler and server are owned by the same process (20519) so Table uniform_table is accessed directly without gRPC.
[reverb/cc/client.cc:165] Sampler and server are owned by the same process (20519) so Table uniform_table is accessed directly without gRPC.
[reverb/cc/client.cc:165] Sampler and server are owned by the same process (20519) so Table uniform_table is accessed directly without gRPC.
[reverb/cc/client.cc:165] Sampler and server are owned by the same process (20519) so Table uniform_table is accessed directly without gRPC.
[reverb/cc/client.cc:165] Sampler and server are owned by the same process (20519) so Table uniform_table is accessed directly without gRPC.
[reverb/cc/client.cc:165] Sampler and server are owned by the same process (20519) so Table uniform_table is accessed directly without gRPC.
step = 200: loss = 20.331722259521484
step = 400: loss = 1.3115228414535522
step = 600: loss = 3.6862149238586426
step = 800: loss = 25.54358673095703
step = 1000: loss = 93.734619140625
step = 1000: Average Return = 46.400001525878906
step = 1200: loss = 24.599149703979492
step = 1400: loss = 60.56067657470703
step = 1600: loss = 760.88037109375
step = 1800: loss = 106.09115600585938
step = 2000: loss = 770.9517822265625
step = 2000: Average Return = 200.0
step = 2200: loss = 232.77313232421875
step = 2400: loss = 954.6015625
step = 2600: loss = 7539.4580078125
step = 2800: loss = 471.50140380859375
step = 3000: loss = 421.0692443847656
step = 3000: Average Return = 200.0
step = 3200: loss = 247.37408447265625
step = 3400: loss = 640.3017578125
step = 3600: loss = 15842.451171875
step = 3800: loss = 2536.94091796875
step = 4000: loss = 963.9180908203125
step = 4000: Average Return = 200.0
step = 4200: loss = 423.5577697753906
step = 4400: loss = 403.044921875
step = 4600: loss = 954.7439575195312
step = 4800: loss = 289.0184631347656
step = 5000: loss = 675.5223388671875
step = 5000: Average Return = 200.0
step = 5200: loss = 789.85888671875
step = 5400: loss = 39104.31640625
step = 5600: loss = 935.078125
step = 5800: loss = 70645.5
step = 6000: loss = 81613.0546875
step = 6000: Average Return = 200.0
step = 6200: loss = 670.2115478515625
step = 6400: loss = 312.8699951171875
step = 6600: loss = 484.3177490234375
step = 6800: loss = 589.077880859375
step = 7000: loss = 633.8604125976562
step = 7000: Average Return = 200.0
step = 7200: loss = 399.37847900390625
step = 7400: loss = 866.277099609375
step = 7600: loss = 491.73382568359375
step = 7800: loss = 372.16693115234375
step = 8000: loss = 254.13558959960938
step = 8000: Average Return = 200.0
step = 8200: loss = 401.8952331542969
step = 8400: loss = 223.53965759277344
step = 8600: loss = 422.3853759765625
step = 8800: loss = 1538.7784423828125
step = 9000: loss = 597.556396484375
step = 9000: Average Return = 200.0
step = 9200: loss = 618.9515991210938
step = 9400: loss = 561.2392578125
step = 9600: loss = 491.34844970703125
step = 9800: loss = 229.70156860351562
step = 10000: loss = 13431.5908203125
step = 10000: Average Return = 200.0
step = 10200: loss = 292.134521484375
step = 10400: loss = 804.981689453125
step = 10600: loss = 197.18800354003906
step = 10800: loss = 606.76416015625
step = 11000: loss = 440.2492370605469
step = 11000: Average Return = 200.0
step = 11200: loss = 303.9647216796875
step = 11400: loss = 113.16478729248047
step = 11600: loss = 280.0725402832031
step = 11800: loss = 115.59662628173828
step = 12000: loss = 125.70242309570312
step = 12000: Average Return = 200.0
step = 12200: loss = 218.43441772460938
step = 12400: loss = 1094.86962890625
step = 12600: loss = 213.63992309570312
step = 12800: loss = 356.6697998046875
step = 13000: loss = 825.29736328125
step = 13000: Average Return = 200.0
step = 13200: loss = 357.39111328125
step = 13400: loss = 528.3963623046875
step = 13600: loss = 319.46343994140625
step = 13800: loss = 510.74554443359375
step = 14000: loss = 496.3047790527344
step = 14000: Average Return = 200.0
step = 14200: loss = 1039.976806640625
step = 14400: loss = 1224.116943359375
step = 14600: loss = 150.0677490234375
step = 14800: loss = 47494.9140625
step = 15000: loss = 1130.8258056640625
step = 15000: Average Return = 200.0
step = 15200: loss = 1225.6185302734375
step = 15400: loss = 1091.06640625
step = 15600: loss = 1100.02099609375
step = 15800: loss = 1358.365478515625
step = 16000: loss = 525.7699584960938
step = 16000: Average Return = 200.0
step = 16200: loss = 1316.208251953125
step = 16400: loss = 1414.6695556640625
step = 16600: loss = 984.55029296875
step = 16800: loss = 1629.3614501953125
step = 17000: loss = 1256.438232421875
step = 17000: Average Return = 200.0
step = 17200: loss = 180060.078125
step = 17400: loss = 1382.09912109375
step = 17600: loss = 94595.5703125
step = 17800: loss = 1473.7625732421875
step = 18000: loss = 9914.451171875
step = 18000: Average Return = 200.0
step = 18200: loss = 1026.8931884765625
step = 18400: loss = 861.779541015625
step = 18600: loss = 1655.6065673828125
step = 18800: loss = 1006.54150390625
step = 19000: loss = 724.9564819335938
step = 19000: Average Return = 200.0
step = 19200: loss = 184484.34375
step = 19400: loss = 613.6680908203125
step = 19600: loss = 3162.27001953125
step = 19800: loss = 1011.3887329101562
step = 20000: loss = 3129.92919921875
step = 20000: Average Return = 200.0

Visualization

Plots

Use matplotlib.pyplot to chart how the policy improved during training.

One iteration of Cartpole-v0 consists of 200 time steps. The environment gives a reward of +1 for each step the pole stays up, so the maximum return for one episode is 200. The charts shows the return increasing towards that maximum each time it is evaluated during training. (It may be a little unstable and not increase monotonically each time.)

iterations = range(0, num_iterations + 1, eval_interval)
plt.plot(iterations, returns)
plt.ylabel('Average Return')
plt.xlabel('Iterations')
plt.ylim(top=250)

(0.1849997997283932, 250.0)

png

Videos

Charts are nice. But more exciting is seeing an agent actually performing a task in an environment.

First, create a function to embed videos in the notebook.

def embed_mp4(filename):
  """Embeds an mp4 file in the notebook."""
  video = open(filename,'rb').read()
  b64 = base64.b64encode(video)
  tag = '''
  <video width="640" height="480" controls>
    <source src="data:video/mp4;base64,{0}" type="video/mp4">
  Your browser does not support the video tag.
  </video>'''.format(b64.decode())

  return IPython.display.HTML(tag)

Now iterate through a few episodes of the Cartpole game with the agent. The underlying Python environment (the one "inside" the TensorFlow environment wrapper) provides a render() method, which outputs an image of the environment state. These can be collected into a video.

def create_policy_eval_video(policy, filename, num_episodes=5, fps=30):
  filename = filename + ".mp4"
  with imageio.get_writer(filename, fps=fps) as video:
    for _ in range(num_episodes):
      time_step = eval_env.reset()
      video.append_data(eval_py_env.render())
      while not time_step.is_last():
        action_step = policy.action(time_step)
        time_step = eval_env.step(action_step.action)
        video.append_data(eval_py_env.render())
  return embed_mp4(filename)

create_policy_eval_video(agent.policy, "trained-agent")

WARNING:root:IMAGEIO FFMPEG_WRITER WARNING: input image is not divisible by macro_block_size=16, resizing from (400, 600) to (400, 608) to ensure video compatibility with most codecs and players. To prevent resizing, make your input image divisible by the macro_block_size or set the macro_block_size to None (risking incompatibility). You may also see a FFMPEG warning concerning speedloss due to data not being aligned.
[swscaler @ 0x565325a66880] Warning: data is not aligned! This can lead to a speed loss

For fun, compare the trained agent (above) to an agent moving randomly. (It does not do as well.)

create_policy_eval_video(random_policy, "random-agent")

WARNING:root:IMAGEIO FFMPEG_WRITER WARNING: input image is not divisible by macro_block_size=16, resizing from (400, 600) to (400, 608) to ensure video compatibility with most codecs and players. To prevent resizing, make your input image divisible by the macro_block_size or set the macro_block_size to None (risking incompatibility). You may also see a FFMPEG warning concerning speedloss due to data not being aligned.
[swscaler @ 0x555dd7ee2880] Warning: data is not aligned! This can lead to a speed loss