This repository shares my learning journey in the field of Reinforcement Learning.
Reinforcement Learning is a type of machine learning which allows an AI agent to learn from it's surrounding (or environment) by interacting with it. It is kind of a trail and error learning. Each successful action leads to a positive reward, or a negative reward in another case. The agent goes through a series of actions and gains an overall reward for the whole duration of interacting with the environment.
The OpenAI gym provides many standard environments for people to test their Reinforcement Learning algorithms.
All reinforcement learning algorithms tend to maximize the reward over its whole time of learning. But how an agent chooses an action to maximize the reward varies differently and there are several approaches.
Though the idea seems quite intuitive, in practice there are many challenges. It's imperative to know which of the earlier actions were actually responsible for getting the reward and to what extent. This is called the Credit-Assignment Problem. Once you have figured out a strategy to collect a certain number of rewards, should we stick with it or experiment with something that could result in even bigger rewards.
We need to successfully balance the exploration and exploitation tradeoff. Ideally, an approach should encourage exploration until the point it has learned enough about it to make informed decisions in the environment by taking optimal actions.
In order for an agent to learn how to deal optimally with all possible states in an environment, it must be exposed to as many of those states as possible. This can be thought of as exploring all the possible states in any environment. Once an agent has explored enough, it can exploit its experience to maximize the final reward.
Hence, an agent, in this case, explores and then exploits. But this is not the case always.
Some of the common methods of exploration are listed out below along with their implementation. (Thanks to Arthur)
It is a naive method which simply chooses that optimal action which leads to greatest reward from the agent.
Taking the action which the agent estimates at the current moment to be the best is an example of exploitation: the agent is exploiting it's current knowledge about the reward structure of the environment to act.
Implementation:
# Q_out is the acivation from the final layer of Q-network.
Q_value = sess.run(Q_out, feed_dict={inputs:[state]})
action = np.argmax(Q_value)
In the above figure, each value corresponds to the Q-value for a given action at a random state in an environment. The height of the light blue bar corresponds to the probability of choosing a given action. The dark blue bar corresponds to a chosen action.
The opposite to the greedy approach is to always take a random action.
Implementation:
# assuming OpenAI gym environment
action = env.action_space.sample()
action = np.random.randint(0, total_actions)
In the above figure, each value corresponds to the Q-value for a given action at a random state in an environment. The height of the light blue bar corresponds to the probability of choosing a given action. The dark blue bar corresponds to a chosen action.
A simple combination of the greedy and random approaches yields one of the most used exploration strategies: ϵ-greedy. In this approach the agent chooses what it believes to be the optimal action most of the time, but occasionally acts randomly. This way the agent takes actions which it may not estimate to be ideal, but may provide new information to the agent. The ϵ in ϵ-greedy is an adjustable parameter which determines the probability of taking a random, rather than principled, action. Due to its simplicity and surprising power, this approach has become the defacto technique for most recent reinforcement learning algorithms, including DQN and its variants.
Implementation:
e = 0.1
if np.random.randint(1)< e:
action = env.action_space.sample()
else:
Q_dist = sess.run(Q_out,feed_dict={inputs:[state]})
action = np.argmax(Q_dist)
In the above figure, each value corresponds to the Q-value for a given action at a random state in an environment. The height of the light blue bar corresponds to the probability of choosing a given action. The dark blue bar corresponds to a chosen action.
Q-learning attempts to solve the credit assignment problem – it propagates rewards back in time, until it reaches the crucial decision point which was the actual cause for the obtained reward. Touching on the exploration-exploitation dilemma, we firstly observe, that when a Q-table or Q-network is initialized randomly, then its predictions are initially random as well. If we pick an action with the highest Q-value, the action will be random and the agent performs crude “exploration”. As a Q-function converges, it returns more consistent Q-values and the amount of exploration decreases. So one could say, that Q-learning incorporates the exploration as part of the algorithm. But this exploration is “greedy”, it settles with the first effective strategy it finds.
A simple and effective fix for the above problem is ε-greedy exploration – with probability ε choose a random action, otherwise go with the “greedy” action with the highest Q-value. DeepMind in their system actually decreases ε over time from 1 to 0.1 – in the beginning the system makes completely random moves to explore the state space maximally, and then it settles down to a fixed exploration rate.
We generalize any reinforcement learning problem in its simplest form as Markov Decision Process (MDP).
Consider a learning agent, in an environment (eg., Breakout Game). The environment is in a certain state(location of the paddle, location and direction of the ball, existence of every brick and so on).
The agent can perform certain actions in the environment (eg., move the paddle to left or right). These actions sometimes result in a reward (eg., increase in score). Actions transform the environment and lead to a new state, where the agent can perform another action, and so on. The rules for how you are going to choose these actions are called policy. The environment is general is stochastic, which means the next state is somewhat random (rg., when you lose a ball and start afresh).
The set of states and actions, together with rules for transitioning from one state to another and for getting rewards, make up a Markov decision process. One episode of this process (e.g. one game) forms a finite sequence of states, actions and rewards:
s0,a0,r1,s1,a1,r2,s2,…,sn−1,an−1,rn,sn
Here si represents the state, ai is the action and ri+1 is the reward after performing the action. The episode ends with terminal state sn (e.g. “game over” screen). A Markov decision process relies on the Markov assumption, that the probability of the next state si+1 depends only on current state si and performed action ai, but not on preceding states or actions.
To perform well in the long term, we need to not only take into account the immediate awards but also the future rewards we are going to get.
Given one run of Markov Decision Process, we can calcualte the total reward for one episode:
R=r1+r2+r3+…+rn
Given that, the total future reward from time point t onward can be expressed as:
Rt=rt+rt+1+rt+2+…+rn
But because our environment is stochastic, we can never be sure, if we will get the same rewards the next time we perform the same actions. The more into the future we go, the more it may diverge. For that reason it is common to use discounted future reward instead:
Rt=rt+γrt+1+γ2rt+2…+γn−trn
Here γ is the discount factor between 0 and 1 – the more into the future the reward is, the less we take it into consideration. It is easy to see, that discounted future reward at time step t can be expressed in terms of the same thing at time step t+1:
Rt=rt+γ(rt+1+γ(rt+2+…))=rt+γRt+1
If we set the discount factor γ=0, then our strategy will be short-sighted and we rely only on the immediate rewards. If we want to balance between immediate and future rewards, we should set discount factor to something like γ=0.9. If our environment is deterministic and the same actions always result in same rewards, then we can set discount factor γ=1.
A good strategy for an agent would be to always choose an action, that maximizes the discounted future reward.
Neural Networks are the agent tha tries to learn the mapping from state-action pairs to rewards. In reinforcement learning, CNN can be used to recognize an agent's state, eg., the screen that mario is on or the terrain before a drone. They typically perform the image recognition task but, differently here.
In normal supervised learning, they map names to pixels. It will rank the labels that best fit the image in terms of their probabilities.
Here in RL, they derive different representations from images. Given an image that represents a state, a convolutional net can rank the actions possible to perform in that state, for eg., it might predict that jumping will score them 7 points rather than running right.
The above image illustrates what a policy agent does, mapping a state to the best action.
A pole is attached by an un-actuated joint to a cart, which moves along a frictionless track. The system is controlled by applying a force of +1 or -1 to the cart. The pendulum starts upright, and the goal is to prevent it from falling over. A reward of +1 is provided for every timestep that the pole remains upright. The episode ends when the pole is more than 15 degrees from vertical, or the cart moves more than 2.4 units from the center.
In simple terms, we will code out an algorithm to teach our machine how to balance a pole at the top of a cart.
Source : https://gym.openai.com/envs/CartPole-v1/
It's very easy to interact with any environment in the OpenAI gym and is as simple as defining:
env = gym.make('Cartpole-v1')
You can then see the environment using env.render().
Make sure to have gym installed in your system. (pip install gym)
- episodes - the number of games we want the agent to play.
- gamma - aka decay or discount rate, to calculate the future discounted reward.
- epsilon - aka exploration rate, this is the rate in which an agent randomly decides its action rather than prediction.
- epsilon_decay - we want to decrease the number of explorations as it gets good at playing games.
- epsilon_min - we want the agent to explore at least this amount.
- learning_rate - Determines how much a neural net learns in each iteration.
Below is how the training process looks like:
The agent controls the movement of a character in a grid world. Some tiles of the grid are walkable, and others lead to the agent falling into the water. Additionally, the movement direction of the agent is uncertain and only partially depends on the chosen direction.The agent is rewarded for finding a walkable path to a goal tile.
Winter is here. You and your friends were tossing around a frisbee at the park when you made a wild throw that left the frisbee out in the middle of the lake. The water is mostly frozen, but there are a few holes where the ice has melted. If you step into one of those holes, you'll fall into the freezing water. At this time, there's an international frisbee shortage, so it's absolutely imperative that you navigate across the lake and retrieve the disc. However, the ice is slippery, so you won't always move in the direction you intend.
The surface is described using a grid like the following:
SFFF (S: starting point, safe)
FHFH (F: frozen surface, safe)
FFFH (H: hole, fall to your doom)
HFFG (G: goal, where the frisbee is located)
The episode ends when you reach the goal or fall in a hole. You receive a reward of 1 if you reach the goal and zero otherwise.
Hence, we need an algorithm which learns long term expected rewards.
The Q-Learning algorithm, in its simplest implementation, is a table of values for every state(row) and action(column) in an environment. Within each cell of the table, we learn how good it is for an agent to take a given action within a given state. In FrozenLake environment, we have 16 possible states(one for each block), and 4 possible actions(the four directions of movement), giving us a 16x4 table of values.
We start by initializing the table uniformly with all zeros, and then as we observe the rewards we obtain for various actions, we update the table accordingly. We make updates to our Q-table using the Bellman equation, which states that the expected long-term reward for a given action is equal to the immediate reward from the current action combined with the expected reward from the best future action taken at the following state.
- episodes - the number of games we want the agent to play.
- gamma - aka decay or discount rate, to calculate the future discounted reward.
- epsilon - aka exploration rate, this is the rate in which an agent randomly decides its action rather than prediction.
- epsilon_decay - we want to decrease the number of explorations as it gets good at playing games.
- epsilon_min - we want the agent to explore at least this amount.
- A monster is spawned randomly somewhere along the opposite wall.
- The player can only go left/right and shoot.
- 1 hit is enough to kill the monster.
- Episode finishes when the monster is killed or on timeout (300).
REWARDS:
- +101 for killing the monster.
- -5 for missing.
- The episode ends after killing the monster or on timeout.
- living reward = -1
Use this to install doom environment in your machine.
- episodes - the number of games we want the agent to play.
- gamma - aka decay or discount rate, to calculate the future discounted reward.
- epsilon - aka exploration rate, this is the rate in which an agent randomly decides its action rather than prediction.
- epsilon_decay - we want to decrease the number of explorations as it gets good at playing games.
- epsilon_min - we want the agent to explore at least this amount.
- learning_rate - Determines how much a neural net learns in each iteration.