This notebook implements a basic reinforce algorithm a.k.a. policy gradient for CartPole env.
It has been deliberately written to be as simple and human-readable.
Authors: Practical_RL course team
The notebook assumes that you have openai gym installed.
In case you're running on a server, use xvfb
In [1]:
import gym
import numpy as np, pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
env = gym.make("CartPole-v0")
#gym compatibility: unwrap TimeLimit
if hasattr(env,'env'):
env=env.env
env.reset()
n_actions = env.action_space.n
state_dim = env.observation_space.shape
plt.imshow(env.render("rgb_array"))
Out[1]:
For REINFORCE algorithm, we'll need a model that predicts action probabilities given states.
In [2]:
import tensorflow as tf
import sonnet as snt
#create input variables. We only need <s,a,R> for REINFORCE
states = tf.placeholder('float32',(None,)+state_dim,name="states")
actions = tf.placeholder('int32',name="action_ids")
cumulative_rewards = tf.placeholder('float32', name="cumulative_returns")
In [3]:
def make_network(inputs):
lin1 = snt.Linear(output_size=100)(inputs)
elu1 = tf.nn.elu(lin1)
logits = snt.Linear(output_size=n_actions)(elu1)
policy = tf.nn.softmax(logits)
log_policy = tf.nn.log_softmax(logits)
return logits, policy, log_policy
net = snt.Module(make_network,name="policy_network")
logits,policy,log_policy = net(states)
In [4]:
#utility function to pick action in one given state
get_action_proba = lambda s: policy.eval({states:[s]})[0]
We now need to define objective and update over policy gradient.
The objective function can be defined thusly:
$$ J \approx \sum _i log \pi_\theta (a_i | s_i) \cdot R(s_i,a_i) $$When you compute gradient of that function over network weights $ \theta $, it will become exactly the policy gradient.
In [5]:
#REINFORCE objective function
actions_1hot = tf.one_hot(actions,n_actions)
log_pi_a = -tf.nn.softmax_cross_entropy_with_logits(logits=logits,labels=actions_1hot)
J = tf.reduce_mean(log_pi_a * cumulative_rewards)
In [6]:
#regularize with entropy
entropy = -tf.reduce_mean(policy*log_policy)
In [7]:
#all network weights
all_weights = net.get_variables()
#weight updates. maximizing J is same as minimizing -J
loss = -J -0.1*entropy
update = tf.train.AdamOptimizer().minimize(loss,var_list=all_weights)
In [8]:
def get_cumulative_rewards(rewards, #rewards at each step
gamma = 0.99 #discount for reward
):
"""
take a list of immediate rewards r(s,a) for the whole session
compute cumulative rewards R(s,a) (a.k.a. G(s,a) in Sutton '16)
R_t = r_t + gamma*r_{t+1} + gamma^2*r_{t+2} + ...
The simple way to compute cumulative rewards is to iterate from last to first time tick
and compute R_t = r_t + gamma*R_{t+1} recurrently
You must return an array/list of cumulative rewards with as many elements as in the initial rewards.
"""
cumulative_rewards = []
R = 0
for r in rewards[::-1]:
R = r + gamma*R
cumulative_rewards.insert(0,R)
return cumulative_rewards
In [9]:
assert len(get_cumulative_rewards(range(100))) == 100
assert np.allclose(get_cumulative_rewards([0,0,1,0,0,1,0],gamma=0.9),[1.40049, 1.5561, 1.729, 0.81, 0.9, 1.0, 0.0])
assert np.allclose(get_cumulative_rewards([0,0,1,-2,3,-4,0],gamma=0.5), [0.0625, 0.125, 0.25, -1.5, 1.0, -4.0, 0.0])
assert np.allclose(get_cumulative_rewards([0,0,1,2,3,4,0],gamma=0), [0, 0, 1, 2, 3, 4, 0])
print("looks good!")
In [10]:
def train_step(_states,_actions,_rewards):
"""given full session, trains agent with policy gradient"""
_cumulative_rewards = get_cumulative_rewards(_rewards)
update.run({states:_states,actions:_actions,cumulative_rewards:_cumulative_rewards})
In [11]:
def generate_session(t_max=1000):
"""play env with REINFORCE agent and train at the session end"""
#arrays to record session
states,actions,rewards = [],[],[]
s = env.reset()
for t in range(t_max):
#action probabilities array aka pi(a|s)
action_probas = get_action_proba(s)
a = np.random.choice(n_actions,p=action_probas)
new_s,r,done,info = env.step(a)
#record session history to train later
states.append(s)
actions.append(a)
rewards.append(r)
s = new_s
if done: break
train_step(states,actions,rewards)
return sum(rewards)
In [ ]:
s = tf.InteractiveSession()
s.run(tf.global_variables_initializer())
for i in range(100):
rewards = [generate_session() for _ in range(100)] #generate new sessions
print ("mean reward:%.3f"%(np.mean(rewards)))
if np.mean(rewards) > 300:
print ("You Win!")
break
In [ ]:
#record sessions
import gym.wrappers
env = gym.wrappers.Monitor(gym.make("CartPole-v0"),directory="videos",force=True)
sessions = [generate_session() for _ in range(100)]
env.close()
In [ ]:
#show video
from IPython.display import HTML
import os
video_names = list(filter(lambda s:s.endswith(".mp4"),os.listdir("./videos/")))
HTML("""
<video width="640" height="480" controls>
<source src="{}" type="video/mp4">
</video>
""".format("./videos/"+video_names[-1])) #this may or may not be _last_ video. Try other indices
In [ ]:
#That's all, thank you for your attention!