%matplotlib inline
Reinforcement Learning (DQN) TutorialΒΆ
Based on the tutorial by:
Author: Adam Paszke https://github.com/apaszke
This tutorial shows how to use PyTorch to train a Deep Q Learning (DQN) agent
on the CartPole-v0 task from the OpenAI Gym https://gym.openai.com/
Task
You can find an official leaderboard with various algorithms and visualizations at the
Gym website https://gym.openai.com/envs/CartPole-v0
The player to decide between two actions - moving the cart left or right - so that the pole attached to it stays upright.
In this task, rewards are:
- +1 for every incremental timestep
- and the environment terminates if
- the pole falls over too far
- or the cart moves more then 2.4 units away from center.
This means better performing scenarios will run for longer duration, accumulating larger return.
Neural networks can solve the task purely by looking at the scene.
- we'll use a patch of the screen centered on the cart as the observation of the current state
- our actions are move left or move right
Strictly speaking, we will present the state as the difference between the current screen patch and the previous one. This will allow the agent to take the velocity of the pole into account from one image.
Packages
First, let's import needed packages. Firstly, we need
gym https://gym.openai.com/docs for the environment
(Install using pip install gym).
We'll also use the following from PyTorch:
- neural networks (
torch.nn) - optimization (
torch.optim) - automatic differentiation (
torch.autograd) - utilities for vision tasks (
torchvision- a separate package https://github.com/pytorch/vision).
!pip install gym[atari]
!pip install pyglet==1.5.0
#!apt-get install python-opengl -y
#!pip install PyOpenGL
#!pip install PyOpenGL_accelerate
!pip install pyvirtualdisplay
import gym
import math
import random
import numpy as np
import matplotlib
import matplotlib.pyplot as plt
from collections import namedtuple, deque
from itertools import count
from PIL import Image
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torchvision.transforms as T
# to display things
import os
from pyvirtualdisplay import Display
from matplotlib import animation , rc
display = Display(visible=0, size=(1400, 900))
display.start()
os.environ["DISPLAY"] = ":" + str(display.display) + "." + str(display._obj._screen)
# setup the environment
env = gym.make('CartPole-v0').unwrapped
# set up matplotlib
is_ipython = 'inline' in matplotlib.get_backend()
if is_ipython:
from IPython import display
plt.ion()
# if gpu is to be used
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
Display the game environmentΒΆ
env.reset()
plt.imshow(env.render('rgb_array'))
plt.grid(False)
frame = []
env.reset()
total_reward = 0
for i in range(100):
action = env.action_space.sample()
state, reward, done, info = env.step(action)
total_reward += reward
img = plt.imshow(env.render('rgb_array'))
frame.append([img])
if done:
break
print("Game terminated after", len(frame), " steps with reward ", total_reward)
fig = plt.figure()
anim = animation.ArtistAnimation(fig, frame, interval=100, repeat_delay=1000, blit=True)
rc('animation', html='jshtml')
anim
Replay MemoryΒΆ
We'll be using experience replay memory for training our DQN. It stores the transitions that the agent observes, allowing us to reuse this data later. By sampling from it randomly, the transitions that build up a batch are decorrelated. It has been shown that this greatly stabilizes and improves the DQN training procedure.
For this, we're going to need two classses:
Transition- a named tuple representing a single transition in our environment. It essentially maps (state, action) pairs to their (next_state, reward) result, with the state being the screen difference image as described later on.ReplayMemory- a cyclic buffer of bounded size that holds the transitions observed recently. It also implements a.sample()method for selecting a random batch of transitions for training.
# the structure of the transition that we store
Transition = namedtuple('Transition', ('state', 'action', 'next_state', 'reward'))
# stores the Experience Replay buffer
class ReplayMemory(object):
def __init__(self, capacity):
self.cap = capacity
self.memory = deque([],maxlen=capacity)
def push(self, *args):
self.memory.append(Transition(*args))
def sample(self, batch_size):
return random.sample(self.memory, batch_size)
def __len__(self):
return len(self.memory)
Now, let's define our model. But first, let's quickly recap what a DQN is.
DQN algorithmΒΆ
Our environment is deterministic, so all equations presented here are also formulated deterministically for the sake of simplicity. In the reinforcement learning literature, they would also contain expectations over stochastic transitions in the environment.
Our aim will be to train a policy that tries to maximize the discounted, cumulative reward , where is also known as the return. The discount, , should be a constant between and that ensures the sum converges. It makes rewards from the uncertain far future less important for our agent than the ones in the near future that it can be fairly confident about.
The main idea behind Q-learning is that if we had a function , that could tell us what our return would be, if we were to take an action in a given state, then we could easily construct a policy that maximizes our rewards:
However, we don't know everything about the world, so we don't have access to . But, since neural networks are universal function approximators, we can simply create one and train it to resemble .
For our training update rule, we'll use a fact that every function for some policy obeys the Bellman equation:
The difference between the two sides of the equality is known as the temporal difference error, :
To minimise this error, we will use the Smooth L1 Loss aka Huber loss https://en.wikipedia.org/wiki/Huber_loss. The Huber loss acts like the mean squared error when the error is small, but like the mean absolute error when the error is large - this makes it more robust to outliers when the estimates of are very noisy. We calculate this over a batch of transitions, , sampled from the replay memory:
Q-networkΒΆ
Our model will be a convolutional neural network that takes in the difference between the current and previous screen patches. It has two outputs, representing and (where is the input to the network). In effect, the network is trying to predict the expected return of taking each action given the current input.
class DQN(nn.Module):
def __init__(self, h, w, outputs):
super(DQN, self).__init__()
self.conv1 = nn.Conv2d(3, 16, kernel_size=5, stride=2)
self.bn1 = nn.BatchNorm2d(16)
self.conv2 = nn.Conv2d(16, 32, kernel_size=5, stride=2)
self.bn2 = nn.BatchNorm2d(32)
self.conv3 = nn.Conv2d(32, 32, kernel_size=5, stride=2)
self.bn3 = nn.BatchNorm2d(32)
# Number of Linear input connections depends on output of conv2d layers
# and therefore the input image size, so compute it.
def conv2d_size_out(size, kernel_size = 5, stride = 2):
return (size - (kernel_size - 1) - 1) // stride + 1
convw = conv2d_size_out(conv2d_size_out(conv2d_size_out(w)))
convh = conv2d_size_out(conv2d_size_out(conv2d_size_out(h)))
linear_input_size = convw * convh * 32
self.head = nn.Linear(linear_input_size, outputs)
# Called with either one element to determine next action, or a batch
# during optimization. Returns tensor([[left0exp,right0exp]...]).
def forward(self, x):
x = x.to(device)
x = F.relu(self.bn1(self.conv1(x)))
x = F.relu(self.bn2(self.conv2(x)))
x = F.relu(self.bn3(self.conv3(x)))
return self.head(x.view(x.size(0), -1))
Input extractionΒΆ
The code below are utilities for extracting and processing rendered
images from the environment. It uses the torchvision package, which
makes it easy to compose image transforms. Once you run the cell it will
display an example patch that it extracted.
resize = T.Compose([T.ToPILImage(),
T.Resize(40, interpolation=Image.CUBIC),
T.ToTensor()])
def get_cart_location(screen_width):
world_width = env.x_threshold * 2
scale = screen_width / world_width
return int(env.state[0] * scale + screen_width / 2.0) # MIDDLE OF CART
def get_screen():
# Returned screen requested by gym is 400x600x3, but is sometimes larger
# such as 800x1200x3. Transpose it into torch order (CHW).
screen = env.render(mode='rgb_array').transpose((2, 0, 1))
# Cart is in the lower half, so strip off the top and bottom of the screen
_, screen_height, screen_width = screen.shape
screen = screen[:, int(screen_height*0.4):int(screen_height * 0.8)]
view_width = int(screen_width * 0.6)
cart_location = get_cart_location(screen_width)
if cart_location < view_width // 2:
slice_range = slice(view_width)
elif cart_location > (screen_width - view_width // 2):
slice_range = slice(-view_width, None)
else:
slice_range = slice(cart_location - view_width // 2,
cart_location + view_width // 2)
# Strip off the edges, so that we have a square image centered on a cart
screen = screen[:, :, slice_range]
# Convert to float, rescale, convert to torch tensor
# (this doesn't require a copy)
screen = np.ascontiguousarray(screen, dtype=np.float32) / 255
screen = torch.from_numpy(screen)
# Resize, and add a batch dimension (BCHW)
return resize(screen).unsqueeze(0)
env.reset()
plt.figure()
plt.imshow(get_screen().cpu().squeeze(0).permute(1, 2, 0).numpy(), interpolation='none')
#plt.imshow(env.render(mode='rgb_array'))
plt.title('Example extracted screen')
plt.show()
TrainingΒΆ
Hyperparameters and utilitiesΒΆ
This cell instantiates our model and its optimizer, and defines some utilities:
select_action- will select an action accordingly to an epsilon greedy policy. Simply put, we'll sometimes use our model for choosing the action, and sometimes we'll just sample one uniformly. The probability of choosing a random action will start atEPS_STARTand will decay exponentially towardsEPS_END.EPS_DECAYcontrols the rate of the decay.plot_durations- a helper for plotting the durations of episodes, along with an average over the last 100 episodes (the measure used in the official evaluations). The plot will be underneath the cell containing the main training loop, and will update after every episode.
def select_action(state, policy=None, train=True):
global steps_done
sample = random.random()
eps_threshold = EPS_END + (EPS_START - EPS_END) * math.exp(-1. * steps_done / EPS_DECAY)
steps_done += 1
with torch.no_grad():
# t.max(1) will return largest column value of each row.
# second column on max result is index of where max element was
# found, so we pick action with the larger expected reward.
action = policy(state).max(1)[1].view(1, 1)
if train:
if sample > eps_threshold:
return action
else:
return torch.tensor([[random.randrange(n_actions)]], device=device, dtype=torch.long)
else:
return action
episode_durations = []
def plot_durations():
plt.figure(2)
plt.clf()
durations_t = torch.tensor(episode_durations, dtype=torch.float)
plt.title('Training...')
plt.xlabel('Episode')
plt.ylabel('Duration')
plt.plot(durations_t.numpy())
# Take 100 episode averages and plot them too
if len(durations_t) >= 100:
means = durations_t.unfold(0, 100, 1).mean(1).view(-1)
means = torch.cat((torch.zeros(99), means))
plt.plot(means.numpy())
plt.pause(0.001) # pause a bit so that plots are updated
if is_ipython:
display.clear_output(wait=True)
display.display(plt.gcf())
Training loopΒΆ
Finally, the code for training our model.
Here, you can find an optimize_model function that performs a
single step of the optimization. It first samples a batch, concatenates
all the tensors into a single one, computes and
, and combines them into our
loss. By definition we set if is a terminal
state. We also use a target network to compute for
added stability. The target network has its weights kept frozen most of
the time, but is updated with the policy network's weights every so often.
This is usually a set number of steps but we shall use episodes for
simplicity.
def optimize_model():
if len(memory) < BATCH_SIZE:
return
transitions = memory.sample(BATCH_SIZE)
# Transpose the batch (see https://stackoverflow.com/a/19343/3343043 for
# detailed explanation). This converts batch-array of Transitions
# to Transition of batch-arrays.
batch = Transition(*zip(*transitions))
# Compute a mask of non-final states and concatenate the batch elements
# (a final state would've been the one after which simulation ended)
non_final_mask = torch.tensor(tuple(map(lambda s: s is not None,
batch.next_state)), device=device, dtype=torch.bool)
non_final_next_states = torch.cat([s for s in batch.next_state
if s is not None])
state_batch = torch.cat(batch.state)
action_batch = torch.cat(batch.action)
reward_batch = torch.cat(batch.reward)
# Compute Q(s_t, a) - the model computes Q(s_t), then we select the
# columns of actions taken. These are the actions which would've been taken
# for each batch state according to policy_net
state_action_values = policy_net(state_batch).gather(1, action_batch)
# Compute V(s_{t+1}) for all next states.
# Expected values of actions for non_final_next_states are computed based
# on the "older" target_net; selecting their best reward with max(1)[0].
# This is merged based on the mask, such that we'll have either the expected
# state value or 0 in case the state was final.
next_state_values = torch.zeros(BATCH_SIZE, device=device)
next_state_values[non_final_mask] = target_net(non_final_next_states).max(1)[0].detach()
# Compute the expected Q values
expected_state_action_values = (next_state_values * GAMMA) + reward_batch
# Compute Huber loss
criterion = nn.SmoothL1Loss()
loss = criterion(state_action_values, expected_state_action_values.unsqueeze(1))
# Optimize the model
optimizer.zero_grad()
loss.backward()
for param in policy_net.parameters():
param.grad.data.clamp_(-1, 1)
optimizer.step()
Below, you can find the main training loop. At the beginning we reset
the environment and initialize the state Tensor. Then, we sample
an action, execute it, observe the next screen and the reward (always
1), and optimize our model once. When the episode ends (our model
fails), we restart the loop.
Below, num_episodes is set small. You should download
the notebook and run lot more epsiodes, such as 300+ for meaningful
duration improvements.
BATCH_SIZE = 128
GAMMA = 0.999
EPS_START = 0.9
EPS_END = 0.05
EPS_DECAY = 200
TARGET_UPDATE = 10
# Get screen size so that we can initialize layers correctly based on shape
# returned from AI gym. Typical dimensions at this point are close to 3x40x90
# which is the result of a clamped and down-scaled render buffer in get_screen()
init_screen = get_screen()
_, _, screen_height, screen_width = init_screen.shape
# Get number of actions from gym action space
n_actions = env.action_space.n
policy_net = DQN(screen_height, screen_width, n_actions).to(device)
target_net = DQN(screen_height, screen_width, n_actions).to(device)
target_net.load_state_dict(policy_net.state_dict())
target_net.eval()
optimizer = optim.RMSprop(policy_net.parameters())
memory = ReplayMemory(10000)
steps_done = 0
num_episodes = 200
for i_episode in range(num_episodes+1):
# Initialize the environment and state
env.reset()
last_screen = get_screen()
current_screen = get_screen()
state = current_screen - last_screen
for t in count():
# Select and perform an action
action = select_action(state, policy_net)
_, reward, done, _ = env.step(action.item())
reward = torch.tensor([reward], device=device)
# Observe new state
last_screen = current_screen
current_screen = get_screen()
if not done:
next_state = current_screen - last_screen
else:
next_state = None
# Store the transition in memory
memory.push(state, action, next_state, reward)
# Move to the next state
state = next_state
# Perform one step of the optimization (on the policy network)
optimize_model()
if done:
episode_durations.append(t + 1)
#plot_durations()
break
# Update the target network, copying all weights and biases in DQN
if i_episode % TARGET_UPDATE == 0:
print("Completed Episode", i_episode)
target_net.load_state_dict(policy_net.state_dict())
if i_episode % 50 == 0:
print("Playing a test game after episode ", i_episode)
frame = []
env.reset()
last_screen = get_screen()
current_screen = get_screen()
state = current_screen - last_screen
total_reward = 0
for i in range(1000):
if i == 0:
action = env.action_space.sample()
action = select_action(state, policy_net, train=False)
pesudo_state, reward, done, info = env.step(action.item())
#reward = torch.tensor([reward], device=device)
total_reward += reward
# Observe new state
last_screen = current_screen
current_screen = get_screen()
if not done:
state = current_screen - last_screen
else:
break
#img = plt.imshow(env.render('rgb_array'))
#frame.append([img])
if done:
break
print("Game terminated after", len(frame), "steps with reward", total_reward)
print('Complete')
env.render()
env.close()
plt.ioff()
plt.show()
Play a gameΒΆ
sum_reward=0
for j in range(100):
env.reset()
frame = []
last_screen = get_screen()
current_screen = get_screen()
state = current_screen - last_screen
total_reward = 0
for i in range(500):
if i == 0:
action = env.action_space.sample()
action = select_action(state, policy_net, train=False)
pesudo_state, reward, done, info = env.step(action.item())
#reward = torch.tensor([reward], device=device)
total_reward += reward
# Observe new state
last_screen = current_screen
current_screen = get_screen()
if not done:
state = current_screen - last_screen
else:
break
#img = plt.imshow(env.render('rgb_array'))
#frame.append([img])
#print(i, action.item())
if done:
break
print("Game ", j , " terminated after", len(frame), "steps with reward", total_reward)
sum_reward += total_reward
print("Average reward", sum_reward/100)
#fig = plt.figure()
#anim = animation.ArtistAnimation(fig, frame, interval=100, repeat_delay=1000, blit=True)
#rc('animation', html='jshtml')
#anim
sum_reward=0
for j in range(100):
env.reset()
frame = []
total_reward = 0
for i in range(500):
action = env.action_space.sample()
pesudo_state, reward, done, info = env.step(action)
total_reward += reward
img = plt.imshow(env.render('rgb_array'))
frame.append([img])
#print(i, action.item())
if done:
break
print("Game ", j , " terminated after", len(frame), "steps with reward", total_reward)
sum_reward += total_reward
print("Average reward", sum_reward/100)
Here is the diagram that illustrates the overall resulting data flow.
Actions are chosen either randomly or based on a policy, getting the next step sample from the gym environment. We record the results in the replay memory and also run optimization step on every iteration. Optimization picks a random batch from the replay memory to do training of the new policy. "Older" target_net is also used in optimization to compute the expected Q values; it is updated occasionally to keep it current.