Pacman seeks reward.
Should he eat or should he run?
When in doubt, Q-learn.
In this project, you will implement value iteration and Q-learning. You will test your agents first on Gridworld (from class), then apply them to a simulated robot controller (Crawler) and Pacman.
The code for this project contains the following files, which are available in a zip archive:
||A value iteration agent for solving known MDPs.|
||Q-learning agents for Gridworld, Crawler and Pacman.|
||A file to put your answers to questions given in the project.|
||Defines methods on general MDPs.|
||Defines the base classes
||The Gridworld implementation.|
||Classes for extracting features on (state,action) pairs. Used for the approximate Q-learning agent (in qlearningAgents.py).|
||Abstract class for general reinforcement learning environments. Used
||Gridworld graphical display.|
||Plug-in for the Gridworld text interface.|
||The crawler code and test harness. You will run this but not edit it.|
||GUI for the crawler robot.|
What to submit: You will fill in portions of
analysis.py during the assignment. You should submit only these files. Please don't change any others.
Evaluation: Your code will be autograded for technical correctness. Please do not change the names of any provided functions or classes within the code, or you will wreak havoc on the autograder. If your code works correctly on one or two of the provided examples but doesn't get full credit from the autograder, you most likely have a subtle bug that breaks one of our more thorough test cases; you will need to debug more fully by reasoning about your code and trying small examples of your own. That said, bugs in the autograder are not impossible, so please do contact the staff if you believe that there has been an error in the grading.
Academic Integrity: We will be checking your code against other submissions in the class for logical redundancy. If you copy someone else's code and submit it with minor changes, we will know. These cheat detectors are quite hard to fool, so please don't try. We trust you all to submit your own work only; please don't let us down. If you do, we will pursue the strongest consequences available to us.
Getting Help: You are not alone! If you find yourself stuck on something, contact the course staff for help. Office hours, section, and Piazza are there for your support; please use them. If you can't make our office hours, let us know and we will schedule more. We want these projects to be rewarding and instructional, not frustrating and demoralizing. But, we don't know when or how to help unless you ask.
To get started, run Gridworld in manual control mode, which uses the arrow keys:
python gridworld.py -m
You will see the two-exit layout from class. The blue dot is the agent. Note that when you press up, the agent only actually moves north 80% of the time. Such is the life of a Gridworld agent!
You can control many aspects of the simulation. A full list of options is available by running:
python gridworld.py -h
The default agent moves randomly
python gridworld.py -g MazeGrid
You should see the random agent bounce around the grid until it happens upon an exit. Not the finest hour for an AI agent.
Note: The Gridworld MDP is such that you first must enter a pre-terminal state (the double boxes shown in the GUI) and then take the special 'exit' action before the episode actually ends (in the true terminal state called
TERMINAL_STATE, which is not shown in the GUI). If you run an episode manually, your total return may be less than you expected, due to the discount rate (
-d to change; 0.9 by default).
Look at the console output that accompanies the graphical output (or use
-t for all text).
You will be told about each transition the agent experiences (to turn this off, use
As in Pacman, positions are represented by
(x,y) Cartesian coordinates
and any arrays are indexed by
the direction of increasing
y, etc. By default,
most transitions will receive a reward of zero, though you can change this
with the living reward option (
Question 1 (6 points) Write a value iteration agent in
ValueIterationAgent, which has been partially specified for you in
valueIterationAgents.py. Your value iteration agent is an offline planner, not a reinforcement learning agent, and so the relevant training option is the number of iterations of value iteration it should run (option
-i) in its initial planning phase.
ValueIterationAgent takes an MDP on construction and runs value iteration for the specified number of iterations before the constructor returns.
Value iteration computes k-step estimates of the optimal values, Vk. In addition to running value iteration, implement the following methods for
ValueIterationAgent using Vk.
getValue(state)returns the value of a state.
getPolicy(state)returns the best action according to computed values.
getQValue(state, action)returns the Q-value of the (state, action) pair.
These quantities are all displayed in the GUI: values are numbers in squares, Q-values are numbers in square quarters, and policies are arrows out from each square.
Important: Use the "batch" version of value iteration where each vector Vk is computed from a fixed vector Vk-1 (like in lecture), not the "online" version where one single weight vector is updated in place. The difference is discussed in Sutton & Barto in the 6th paragraph of chapter 4.1.
Note: A policy synthesized from values of depth k (which reflect the next k rewards) will actually reflect the next k+1 rewards (i.e. you return k+1). Similarly, the Q-values will also reflect one more reward than the values (i.e. you return Qk+1). You may either return the synthesized policy k+1 or the actual policy for the kth iteration, k, which you'll get if you store optimal actions from the most recent round of value iteration updates.
The following command loads your
ValueIterationAgent, which will compute a policy and execute it 10 times. Press a key to cycle through values, Q-values, and the simulation. You should find that the value of the start state (
V(start), which you can read off of the GUI) and the empirical resulting average reward (printed after the 10 rounds of execution finish) are quite close.
python gridworld.py -a value -i 100 -k 10
Hint: On the default BookGrid, running value iteration for 5 iterations should give you this output:
python gridworld.py -a value -i 5
Hint: Use the
util.Counter class in
which is a dictionary with a default value of zero. Methods such as
totalCount should simplify your code. However, be careful with
argMax: the actual argmax you want may be a key not in the counter!
Grading: Your value iteration agent will be graded on a new grid. We will check your values, Q-values, and policies after fixed numbers of iterations and at convergence (e.g. after 100 iterations).
Question 2 (1 point)
BridgeGrid is a grid world map with the a low-reward terminal state and a high-reward terminal state separated by a narrow "bridge", on either side of which is a chasm of high negative reward. The agent starts near the low-reward state. With the default discount of 0.9 and the default noise of 0.2, the optimal policy does not cross the bridge. Change only ONE of the discount and noise parameters so that the optimal policy causes the agent to attempt to cross the bridge. Put your answer in
analysis.py. (Noise refers to how often an agent ends up in an unintended successor state when they perform an action.) The default corresponds to:
python gridworld.py -a value -i 100 -g BridgeGrid --discount 0.9 --noise 0.2
Grading: We will check that you only changed one of the given parameters, and that with this change, a correct value iteration agent should cross the bridge. Note that this part is tested with our value iteration agent, so if you have an error in value iteration, your agent may cross the bridge for incorrect settings of the parameters.
Question 3 (5 points) Consider
DiscountGrid layout, shown below. This grid has two
terminal states with positive payoff (shown in green), a close exit
with payoff +1 and a distant exit with payoff +10. The bottom row of
the grid consists of terminal states with negative payoff (shown in
red); each state in this "cliff" region has payoff -10. The starting
state is the yellow square. We distinguish between two types of
paths: (1) paths that "risk the cliff" and travel near the bottom
row of the grid; these paths are shorter but risk earning a large
negative payoff, and are represented by the red arrow in the figure
below. (2) paths that "avoid the cliff" and travel along the top
edge of the grid. These paths are longer but are less likely to
incur huge negative payoffs. These paths are represented by the
green arrow in the figure below.
In this question, you will choose settings of the discount, noise, and living reward parameters for this MDP to produce optimal policies of several different types. Your setting of the parameter values for each part should have the property that, if your agent followed its optimal policy without being subject to any noise, it would exhibit the given behavior. If a particular behavior is not achieved for any setting of the parameters, assert that the policy is impossible by returning the string
Here are the optimal policy types you should attempt to produce:
You can check your answers by running value iteration on DiscountGrid with specified discount, noise, and living reward parameters as follows:
python gridworld.py -a value -i 100 -g DiscountGrid --discount 0.9 --noise 0.2 --livingReward 0.0
question3e()should each return a 3-item tuple of (discount, noise, living reward) in
Note: You can check your policies in the GUI. For example, using a correct answer to 3(a), the arrow in (0,1) should point east, the arrow in (1,1) should also point east, and the arrow in (2,1) should point north.
Grading: We will check that the desired policy is returned in each case. As in question 2, we test your parameter settings using our value iteration agent.
Note that your value iteration agent does not actually learn from experience. Rather, it ponders its MDP model to arrive at a complete policy before ever interacting with a real environment. When it does interact with the environment, it simply follows the precomputed policy (e.g. it becomes a reflex agent). This distinction may be subtle in a simulated environment like a Gridword, but it's very important in the real world, where the real MDP is not available.
Question 4 (5 points) You will now write a Q-learning agent, which does very little on construction, but instead learns by trial and error from interactions with the environment through its
update(state, action, nextState, reward) method. A stub of a Q-learner is specified in
qlearningAgents.py, and you can select it with the option
'-a q'. For this question, you must implement the
getPolicy, you should break ties randomly for better behavior. The
random.choice() function will help. In a particular state, actions that your agent hasn't seen before still have a Q-value, specifically a Q-value of zero, and if all of the actions that your agent has seen before have a negative Q-value, an unseen action may be optimal.
Important: Make sure that in your
getPolicy functions, you only access Q values by calling
getQValue . This
abstraction will be useful for question 9 when you
getQValue to use features of state-action
pairs rather than state-action pairs directly.
With the Q-learning update in place, you can watch your Q-learner learn under manual control, using the keyboard:
python gridworld.py -a q -k 5 -mRecall that
-kwill control the number of episodes your agent gets to learn. Watch how the agent learns about the state it was just in, not the one it moves to, and "leaves learning in its wake." Hint: to help with debugging, you can turn off noise by using the
--noise 0.0parameter (though this obviously makes Q-learning less interesting). If you manually steer Pacman north and then east along the optimal path for four episodes, you should see the following Q-values:
Grading: We will run your Q-learning agent on an example of our own and check that it learns the same Q-values and policy as our reference implementation when each is presented with the same set of examples.
Question 5 (2 points) Complete your Q-learning agent by implementing epsilon-greedy action selection in
getAction, meaning it chooses random actions an epsilon fraction of the time, and follows its current best Q-values otherwise.
python gridworld.py -a q -k 100Your final Q-values should resemble those of your value iteration agent, especially along well-traveled paths. However, your average returns will be lower than the Q-values predict because of the random actions and the initial learning phase.
You can choose an element from a list uniformly at random by calling the
You can simulate a binary variable with probability
of success by using
util.flipCoin(p), which returns
False with probability
Question 6 (1 points) First, train a completely random Q-learner with the default learning rate on the noiseless BridgeGrid for 50 episodes and observe whether it finds the optimal policy.
python gridworld.py -a q -k 50 -n 0 -g BridgeGrid -e 1Now try the same experiment with an epsilon of 0. Is there an epsilon and a learning rate for which it is highly likely (greater than 99%) that the optimal policy will be learned after 50 iterations?
analysis.pyshould return EITHER a 2-item tuple of
(epsilon, learning rate)OR the string
'NOT POSSIBLE'if there is none. Epsilon is controlled by
-e, learning rate by
Note: Your response should be not depend on the exact tie-breaking mechanism used to choose actions. This means your answer should be correct even if for instance we rotated the entire bridge grid world 90 degrees.
Question 7 (1 point) With no additional code, you should now be able to run a Q-learning crawler robot:
python crawler.pyIf this doesn't work, you've probably written some code too specific to the
GridWorldproblem and you should make it more general to all MDPs.
This will invoke the crawling robot from class using your Q-learner. Play around with the various learning parameters to see how they affect the agent's policies and actions. Note that the step delay is a parameter of the simulation, whereas the learning rate and epsilon are parameters of your learning algorithm, and the discount factor is a property of the environment.
Grading: We give you a point for free here, but play around with the crawler anyway!
Question 8 (1 points) Time to play some Pacman! Pacman will play games in two phases.
In the first phase, training, Pacman will begin to learn about the values of positions and actions.
Because it takes a very long time to learn accurate Q-values even for tiny grids, Pacman's training games
run in quiet mode by default, with no GUI (or console) display. Once Pacman's training is complete,
he will enter testing mode. When testing, Pacman's
self.alpha will be set to 0.0, effectively stopping Q-learning and disabling exploration, in order to allow Pacman to exploit his learned policy. Test games are shown in the GUI by default. Without any code changes you should be able to run Q-learning Pacman for very tiny grids as follows:
python pacman.py -p PacmanQAgent -x 2000 -n 2010 -l smallGridNote that
PacmanQAgentis already defined for you in terms of the
QLearningAgentyou've already written.
PacmanQAgentis only different in that it has default learning parameters that are more effective for the Pacman problem (
epsilon=0.05, alpha=0.2, gamma=0.8). You will receive full credit for this question if the command above works without exceptions and your agent wins at least 80% of the time. The autograder will run 100 test games after the 2000 training games.
Hint: If your
QLearningAgent works for
crawler.py but does not seem to be learning a good policy for Pacman on
smallGrid, it may be because your
getPolicy methods do not in some cases properly consider unseen actions. In particular, because unseen actions have by definition a Q-value of zero, if all of the actions that have been seen have negative Q-values, an unseen action may be optimal. Beware of the argmax function from util.Counter!
Note: If you want to experiment with learning parameters, you can use the option
-a, for example
-a epsilon=0.1,alpha=0.3,gamma=0.7. These values will then be accessible as
self.epsilon, self.gamma and
self.alpha inside the agent.
Note: While a total of 2010 games will be played, the first 2000 games will not be displayed because of the option
-x 2000, which designates the first 2000 games for training (no output). Thus, you will only see Pacman play the last 10 of these games. The number of training games is also passed to your agent as the option
Note: If you want to watch 10 training games to see what's going on, use the command:
python pacman.py -p PacmanQAgent -n 10 -l smallGrid -a numTraining=10
Make sure you understand what is happening here: the MDP state is the exact board configuration facing Pacman, with the now complex transitions describing an entire ply of change to that state. The intermediate game configurations in which Pacman has moved but the ghosts have not replied are not MDP states, but are bundled in to the transitions.
Once Pacman is done training, he should win very reliably in test games (at least 90% of the time), since now he is exploiting his learned policy.
However, you will find that training the same agent on the seemingly simple
mediumGrid does not work well. In our implementation, Pacman's average training rewards remain negative throughout training. At test time, he plays badly, probably losing all of his test games. Training will also take a long time, despite its ineffectiveness.
Pacman fails to win on larger layouts because each board configuration is a separate state with separate Q-values. He has no way to generalize that running into a ghost is bad for all positions. Obviously, this approach will not scale.
Question 9 (3 points)
You don't have to do the last question because we won't cover function approximation in this class.
qlearningAgents.py, which is a subclass of
Note: Approximate Q-learning assumes the existence of a feature function f(s,a) over state and action pairs, which yields a vector f1(s,a) .. fi(s,a) .. fn(s,a) of feature values. We provide feature functions for you in
featureExtractors.py. Feature vectors are
util.Counter (like a dictionary) objects containing the non-zero pairs of features and values; all omitted features have value zero.
The approximate Q-function takes the following form
ApproximateQAgent uses the
IdentityExtractor, which assigns a single feature to every
(state,action) pair. With this feature extractor, your approximate Q-learning agent should work identically to
PacmanQAgent. You can test this with the following command:
python pacman.py -p ApproximateQAgent -x 2000 -n 2010 -l smallGrid
ApproximateQAgent is a subclass of
QLearningAgent, and it therefore shares several methods like
getAction. Make sure that your methods in
getQValue instead of accessing Q-values directly, so that when you override
getQValue in your approximate agent, the new approximate q-values are used to compute actions.
Once you're confident that your approximate learner works correctly with the identity features, run your approximate Q-learning agent with our custom feature extractor, which can learn to win with ease:
python pacman.py -p ApproximateQAgent -a extractor=SimpleExtractor -x 50 -n 60 -l mediumGridEven much larger layouts should be no problem for your
ApproximateQAgent. (warning: this may take a few minutes to train)
python pacman.py -p ApproximateQAgent -a extractor=SimpleExtractor -x 50 -n 60 -l mediumClassic
If you have no errors, your approximate Q-learning agent should win almost every time with these simple features, even with only 50 training games.
Grading: We will run your agent on the
mediumGrid layout 100 times using a fixed random seed. You will receive 1 point if your agent wins more than 25% of its games, 2 points if it wins more than 50% of its games, and 3 points if it wins more than 75% of its games. You can try your agent out under these conditions with
python pacman.py -p ApproximateQAgent -a extractor=SimpleExtractor -x 50 -n 60 -l mediumGrid -q -f
Congratulations! You have a learning Pacman agent!