Newer
Older
"""Implement Agents and Environments (Chapters 1-2).
The class hierarchies are as follows:
Object ## A physical object that can exist in an environment
Agent
Wumpus
RandomAgent
ReflexVacuumAgent
...
Dirt
Wall
...
Environment ## An environment holds objects, runs simulations
XYEnvironment
VacuumEnvironment
WumpusEnvironment
EnvGUI ## A window with a graphical representation of the Environment
EnvToolbar ## contains buttons for controlling EnvGUI
EnvCanvas ## Canvas to display the environment of an EnvGUI
"""
# TO DO:
# Implement grabbing correctly.
# When an object is grabbed, does it still have a location?
# What if it is released?
# What if the grabbed or the grabber is deleted?
# What if the grabber moves?
#
# Speed control in GUI does not have any effect -- fix it.
from utils import *
import random, copy
# Additional modules needed for loading non-bitmap images
import Image # Python Imaging Library (PIL)
import ImageTk # PIL + Tk
#______________________________________________________________________________
class Object (object):
"""This represents any physical object that can appear in an Environment.
You subclass Object to get the objects you want. Each object can have a
.__name__ slot (used for output only)."""
def __repr__(self):
return '<%s>' % getattr(self, '__name__', self.__class__.__name__)
def is_alive(self):
"""Objects that are 'alive' should return true."""
return hasattr(self, 'alive') and self.alive
def display(self, canvas, x, y, width, height):
"""Display an image of this Object on the canvas."""
pass
def get_image_file (self):
raise NoImageException()
class Agent (Object):
"""An Agent is a subclass of Object with one required slot,
.program, which should hold a function that takes one argument, the
percept, and returns an action. (What counts as a percept or action
will depend on the specific environment in which the agent exists.)
Note that 'program' is a slot, not a method. If it were a method,
then the program could 'cheat' and look at aspects of the agent.
It's not supposed to do that: the program can only look at the
percepts. An agent program that needs a model of the world (and of
the agent itself) will have to build and maintain its own model.
There is an optional slots, .performance, which is a number giving
the performance measure of the agent in its environment."""
def __init__(self):
def program(percept):
return raw_input('Percept=%s; action? ' % percept)
self.program = program
self.alive = True
def can_grab (self, obj):
"""Returns True if this agent can grab this object.
Override for appropriate subclasses of Agent and Object."""
return False
def TraceAgent(agent):
"""Wrap the agent's program to print its input and output. This will let
you see what the agent is doing in the environment."""
old_program = agent.program
def new_program(percept):
action = old_program(percept)
print '%s perceives %s and does %s' % (agent, percept, action)
return action
agent.program = new_program
return agent
#______________________________________________________________________________
class TableDrivenAgent (Agent):
"""This agent selects an action based on the percept sequence.
It is practical only for tiny domains.
To customize it you provide a table to the constructor. [Fig. 2.7]"""
def __init__(self, table):
"Supply as table a dictionary of all {percept_sequence:action} pairs."
## The agent program could in principle be a function, but because
## it needs to store state, we make it a callable instance of a class.
super(TableDrivenAgent, self).__init__()
percepts = []
def program(percept):
percepts.append(percept)
action = table.get(tuple(percepts))
return action
self.program = program
class RandomAgent (Agent):
"An agent that chooses an action at random, ignoring all percepts."
def __init__(self, actions):
super(RandomAgent, self).__init__()
self.program = lambda percept: random.choice(actions)
#______________________________________________________________________________
loc_A, loc_B = (0, 0), (1, 0) # The two locations for the Vacuum world
class ReflexVacuumAgent (Agent):
"A reflex agent for the two-state vacuum environment. [Fig. 2.8]"
def __init__(self):
super(ReflexVacuumAgent, self).__init__()
def program((location, status)):
if status == 'Dirty': return 'Suck'
elif location == loc_A: return 'Right'
elif location == loc_B: return 'Left'
self.program = program
def get_image_file (self): return "images/vacuum.png"
def RandomVacuumAgent():
"Randomly choose one of the actions from the vaccum environment."
return RandomAgent(['Right', 'Left', 'Suck', 'NoOp'])
def TableDrivenVacuumAgent():
"[Fig. 2.3]"
table = {((loc_A, 'Clean'),): 'Right',
((loc_A, 'Dirty'),): 'Suck',
((loc_B, 'Clean'),): 'Left',
((loc_B, 'Dirty'),): 'Suck',
((loc_A, 'Clean'), (loc_A, 'Clean')): 'Right',
((loc_A, 'Clean'), (loc_A, 'Dirty')): 'Suck',
# ...
((loc_A, 'Clean'), (loc_A, 'Clean'), (loc_A, 'Clean')): 'Right',
((loc_A, 'Clean'), (loc_A, 'Clean'), (loc_A, 'Dirty')): 'Suck',
# ...
}
return TableDrivenAgent(table)
class ModelBasedVacuumAgent (Agent):
"An agent that keeps track of what locations are clean or dirty."
def __init__(self):
super(ModelBasedVacuumAgent, self).__init__()
model = {loc_A: None, loc_B: None}
def program((location, status)):
"Same as ReflexVacuumAgent, except if everything is clean, do NoOp"
model[location] = status ## Update the model here
if model[loc_A] == model[loc_B] == 'Clean': return 'NoOp'
elif status == 'Dirty': return 'Suck'
elif location == loc_A: return 'Right'
elif location == loc_B: return 'Left'
self.program = program
#______________________________________________________________________________
class Environment (object):
"""Abstract class representing an Environment. 'Real' Environment classes
inherit from this. Your Environment will typically need to implement:
percept: Define the percept that an agent sees.
execute_action: Define the effects of executing an action.
Also update the agent.performance slot.
The environment keeps a list of .objects and .agents (which is a subset
of .objects). Each agent has a .performance slot, initialized to 0.
Each object has a .location slot, even though some environments may not
need this."""
def __init__(self):
self.objects = []
self.agents = []
def object_classes (self):
return [] ## List of classes that can go into environment
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
def percept(self, agent):
"Return the percept that the agent sees at this point. Override this."
abstract
def execute_action(self, agent, action):
"Change the world to reflect this action. Override this."
abstract
def default_location(self, object):
"Default location to place a new object with unspecified location."
return None
def exogenous_change(self):
"If there is spontaneous change in the world, override this."
pass
def is_done(self):
"By default, we're done when we can't find a live agent."
for agent in self.agents:
if agent.is_alive(): return False
return True
def step(self):
"""Run the environment for one time step. If the
actions and exogenous changes are independent, this method will
do. If there are interactions between them, you'll need to
override this method."""
if not self.is_done():
actions = [agent.program(self.percept(agent))
for agent in self.agents]
for (agent, action) in zip(self.agents, actions):
self.execute_action(agent, action)
self.exogenous_change()
def run(self, steps=1000):
"""Run the Environment for given number of time steps."""
for step in range(steps):
if self.is_done(): return
self.step()
def list_objects_at (self, location, oclass=Object):
"Return all objects exactly at a given location."
return [obj for obj in self.objects
if obj.location == location and isinstance(obj, oclass)]
def some_objects_at (self, location, oclass=Object):
"""Return true if at least one of the objects at location
is an instance of class oclass.
'Is an instance' in the sense of 'isinstance',
which is true if the object is an instance of a subclass of oclass."""
return self.list_objects_at(location, oclass) != []
def add_object(self, obj, location=None):
"""Add an object to the environment, setting its location. Also keep
track of objects that are agents. Shouldn't need to override this."""
obj.location = location or self.default_location(obj)
self.objects.append(obj)
if isinstance(obj, Agent):
obj.performance = 0
self.agents.append(obj)
return self
def delete_object (self, obj):
"""Remove an object from the environment."""
try:
self.objects.remove(obj)
except ValueError, e:
print e
print " in Environment delete_object"
print " Object to be removed: %s at %s" % (obj, obj.location)
trace_list(" from list", self.objects)
if obj in self.agents:
self.agents.remove(obj)
def trace_list (name, objlist):
ol_list = [(obj, obj.location) for obj in objlist]
print "%s: %s" % (name, ol_list)
class XYEnvironment (Environment):
"""This class is for environments on a 2D plane, with locations
labelled by (x, y) points, either discrete or continuous.
Agents perceive objects within a radius. Each agent in the
environment has a .location slot which should be a location such
as (0, 1), and a .holding slot, which should be a list of objects
that are held."""
def __init__(self, width=10, height=10):
super(XYEnvironment, self).__init__()
self.width = width
self.height = height
#update(self, objects=[], agents=[], width=width, height=height)
self.observers = []
def objects_near(self, location, radius):
"Return all objects within radius of location."
radius2 = radius * radius
return [obj for obj in self.objects
if distance2(location, obj.location) <= radius2]
def percept(self, agent):
"By default, agent perceives objects within radius r."
return [self.object_percept(obj, agent)
for obj in self.objects_near(agent)]
def execute_action(self, agent, action):
agent.bump = False
if action == 'TurnRight':
agent.heading = self.turn_heading(agent.heading, -1)
elif action == 'TurnLeft':
agent.heading = self.turn_heading(agent.heading, +1)
elif action == 'Forward':
self.move_to(agent, vector_add(agent.heading, agent.location))
# elif action == 'Grab':
# objs = [obj for obj in self.list_objects_at(agent.location)
# if agent.can_grab(obj)]
# if objs:
# agent.holding.append(objs[0])
elif action == 'Release':
if agent.holding:
agent.holding.pop()
def object_percept(self, obj, agent): #??? Should go to object?
"Return the percept for this object."
return obj.__class__.__name__
def default_location(self, object):
return (random.choice(self.width), random.choice(self.height))
def move_to(self, obj, destination):
"Move an object to a new location."
# Bumped?
obj.bump = self.some_objects_at(destination, Obstacle)
if not obj.bump:
# Move object and report to observers
obj.location = destination
for o in self.observers:
o.object_moved(obj)
def add_object(self, obj, location=(1, 1)):
super(XYEnvironment, self).add_object(obj, location)
obj.holding = []
obj.held = None
# self.objects.append(obj) # done in Environment!
# Report to observers
for obs in self.observers:
obs.object_added(obj)
def delete_object (self, obj):
super(XYEnvironment, self).delete_object(obj)
# Any more to do? Object holding anything or being held?
for obs in self.observers:
obs.object_deleted(obj)
def add_walls(self):
"Put walls around the entire perimeter of the grid."
for x in range(self.width):
self.add_object(Wall(), (x, 0))
self.add_object(Wall(), (x, self.height-1))
for y in range(self.height):
self.add_object(Wall(), (0, y))
self.add_object(Wall(), (self.width-1, y))
def add_observer (self, observer):
"""Adds an observer to the list of observers.
An observer is typically an EnvGUI.
Each observer is notified of changes in move_to and add_object,
by calling the observer's methods object_moved(obj, old_loc, new_loc)
and object_added(obj, loc)."""
self.observers.append(observer)
def turn_heading(self, heading, inc,
headings=[(1, 0), (0, 1), (-1, 0), (0, -1)]):
"Return the heading to the left (inc=+1) or right (inc=-1) in headings."
return headings[(headings.index(heading) + inc) % len(headings)]
class Obstacle (Object):
"""Something that can cause a bump, preventing an agent from
moving into the same square it's in."""
pass
class Wall (Obstacle): pass
#______________________________________________________________________________
## Vacuum environment
class Dirt (Object):
def get_image_file (self): return "images/dirt.png"
class VacuumEnvironment (XYEnvironment):
"""The environment of [Ex. 2.12]. Agent perceives dirty or clean,
and bump (into obstacle) or not; 2D discrete world of unknown size;
performance measure is 100 for each dirt cleaned, and -1 for
each turn taken."""
def __init__(self, width=10, height=10):
super(VacuumEnvironment, self).__init__(width, height)
self.add_walls()
def object_classes (self):
return [Wall, Dirt, ReflexVacuumAgent, RandomVacuumAgent,
TableDrivenVacuumAgent, ModelBasedVacuumAgent]
def percept(self, agent):
"""The percept is a tuple of ('Dirty' or 'Clean', 'Bump' or 'None').
Unlike the TrivialVacuumEnvironment, location is NOT perceived."""
status = if_(self.some_objects_at(agent.location, Dirt),
'Dirty', 'Clean')
bump = if_(agent.bump, 'Bump', 'None')
return (status, bump)
def execute_action(self, agent, action):
if action == 'Suck':
dirt_list = self.list_objects_at(agent.location, Dirt)
if dirt_list != []:
dirt = dirt_list[0]
self.delete_object(dirt)
else:
super(VacuumEnvironment, self).execute_action(agent, action)
if action != 'Nop':
agent.performance -= 1
class TrivialVacuumEnvironment (Environment):
"""This environment has two locations, A and B. Each can be Dirty
or Clean. The agent perceives its location and the location's
status. This serves as an example of how to implement a simple
Environment."""
def __init__(self):
super(TrivialVacuumEnvironment, self).__init__()
self.status = {loc_A:random.choice(['Clean', 'Dirty']),
loc_B:random.choice(['Clean', 'Dirty'])}
def object_classes (self):
return [Wall, Dirt, ReflexVacuumAgent, RandomVacuumAgent,
TableDrivenVacuumAgent, ModelBasedVacuumAgent]
def percept(self, agent):
"Returns the agent's location, and the location status (Dirty/Clean)."
return (agent.location, self.status[agent.location])
def execute_action(self, agent, action):
"""Change agent's location and/or location's status; track performance.
Score 10 for each dirt cleaned; -1 for each move."""
if action == 'Right':
agent.location = loc_B
agent.performance -= 1
elif action == 'Left':
agent.location = loc_A
agent.performance -= 1
elif action == 'Suck':
if self.status[agent.location] == 'Dirty':
agent.performance += 10
self.status[agent.location] = 'Clean'
def default_location(self, object):
"Agents start in either location at random."
return random.choice([loc_A, loc_B])
#______________________________________________________________________________
class SimpleReflexAgent (Agent):
"""This agent takes action based solely on the percept. [Fig. 2.13]"""
def __init__(self, rules, interpret_input):
super(SimpleReflexAgent, self).__init__()
def program(percept):
state = interpret_input(percept)
rule = rule_match(state, rules)
action = rule.action
return action
self.program = program
class ReflexAgentWithState (Agent):
"""This agent takes action based on the percept and state. [Fig. 2.16]"""
def __init__(self, rules, udpate_state):
super(ReflexAgentWithState, self).__init__()
state, action = None, None
def program(percept):
state = update_state(state, action, percept)
rule = rule_match(state, rules)
action = rule.action
return action
self.program = program
#______________________________________________________________________________
## The Wumpus World
class Gold (Object): pass
class Pit (Object): pass
class Arrow (Object): pass
class Wumpus (Agent): pass
class Explorer (Agent): pass
class WumpusEnvironment(XYEnvironment):
def __init__(self, width=10, height=10):
super(WumpusEnvironment, self).__init__(width, height)
self.add_walls()
def object_classes (self):
return [Wall, Gold, Pit, Arrow, Wumpus, Explorer]
## Needs a lot of work ...
#______________________________________________________________________________
def compare_agents(EnvFactory, AgentFactories, n=10, steps=1000):
"""See how well each of several agents do in n instances of an environment.
Pass in a factory (constructor) for environments, and several for agents.
Create n instances of the environment, and run each agent in copies of
each one for steps. Return a list of (agent, average-score) tuples."""
envs = [EnvFactory() for i in range(n)]
return [(A, test_agent(A, steps, copy.deepcopy(envs)))
for A in AgentFactories]
def test_agent(AgentFactory, steps, envs):
"Return the mean score of running an agent in each of the envs, for steps"
total = 0
for env in envs:
agent = AgentFactory()
env.add_object(agent)
env.run(steps)
total += agent.performance
return float(total)/len(envs)
#_________________________________________________________________________
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
_docex = """
a = ReflexVacuumAgent()
a.program
a.program((loc_A, 'Clean')) ==> 'Right'
a.program((loc_B, 'Clean')) ==> 'Left'
a.program((loc_A, 'Dirty')) ==> 'Suck'
a.program((loc_A, 'Dirty')) ==> 'Suck'
e = TrivialVacuumEnvironment()
e.add_object(TraceAgent(ModelBasedVacuumAgent()))
e.run(5)
## Environments, and some agents, are randomized, so the best we can
## give is a range of expected scores. If this test fails, it does
## not necessarily mean something is wrong.
envs = [TrivialVacuumEnvironment() for i in range(100)]
def testv(A): return test_agent(A, 4, copy.deepcopy(envs))
testv(ModelBasedVacuumAgent)
(7 < _ < 11) ==> True
testv(ReflexVacuumAgent)
(5 < _ < 9) ==> True
testv(TableDrivenVacuumAgent)
(2 < _ < 6) ==> True
testv(RandomVacuumAgent)
(0.5 < _ < 3) ==> True
"""
#______________________________________________________________________________
# GUI - Graphical User Interface for Environments
# If you do not have Tkinter installed, either get a new installation of Python
# (Tkinter is standard in all new releases), or delete the rest of this file
# and muddle through without a GUI.
import Tkinter as tk
class EnvGUI (tk.Tk, object):
def __init__ (self, env, title = 'AIMA GUI', cellwidth=50, n=10):
# Initialize window
super(EnvGUI, self).__init__()
self.title(title)
# Create components
canvas = EnvCanvas(self, env, cellwidth, n)
toolbar = EnvToolbar(self, env, canvas)
for w in [canvas, toolbar]:
w.pack(side="bottom", fill="x", padx="3", pady="3")
class EnvToolbar (tk.Frame, object):
def __init__ (self, parent, env, canvas):
super(EnvToolbar, self).__init__(parent, relief='raised', bd=2)
# Initialize instance variables
self.env = env
self.canvas = canvas
self.running = 0
self.delay = 1.0
# Create buttons and other controls
for txt, cmd in [('Step >', self.env.step), ('Run >>', self.run),
('Stop [ ]', self.stop)]:
tk.Button(self, text=txt, command=cmd).pack(side='left')
tk.Label(self, text='Delay').pack(side='left')
scale = tk.Scale(self, orient='h', from_=0.0, to=10, resolution=0.5,
command=lambda d: setattr(parent, 'delay', d))
scale.set(self.delay)
scale.pack(side='left')
def run(self):
print 'run'
self.running = 1
self.background_run()
def stop(self):
print 'stop'
self.running = 0
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
def background_run(self):
if self.running:
self.env.step()
ms = int(1000 * max(float(self.delay), 0.5))
self.after(ms, self.background_run)
class EnvCanvas (tk.Canvas, object):
def __init__ (self, parent, env, cellwidth, n):
canvwidth = cellwidth * n # (cellwidth + 1 ) * n
canvheight = cellwidth * n # (cellwidth + 1) * n
super(EnvCanvas, self).__init__(parent, background="white",
width=canvwidth, height=canvheight)
# Initialize instance variables
self.env = env
self.cellwidth = cellwidth
self.n = n
# Draw the gridlines
if cellwidth:
for i in range(0, n+1):
self.create_line(0, i*cellwidth, n*cellwidth, i*cellwidth)
self.create_line(i*cellwidth, 0, i*cellwidth, n*cellwidth)
self.pack(expand=1, fill='both')
self.pack()
# Set up object_icon dictionary.
# Each object has an icon mapped in the object_icon dictionary.
# The icon may be a Tk image or any other canvas item,
# typically a "text" if no image is found.
self.object_icon = {}
# Set up image dictionary.
# An image is associated with an image file; multiple objects of the
# same kind use the same image.
# Ugly hack: we need to keep a reference to each ImageTk.PhotoImage,
# or it will be garbage collected. This dictionary maps image files
# that have been opened to their PhotoImage objects
self.images = {}
# Bind canvas events.
#self.bind('<Button-1>', self.user_left) ## What should this do?
#self.bind('<Button-2>', self.user_edit_objects)
self.bind('<Button-3>', self.user_add_object)
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
# Draw existing objects
for obj in env.objects:
self.object_added(obj)
# Observe future new objects and object moves
env.add_observer(self)
def add_object_icon (self, obj):
"""Return a drawable representation for a newly added object obj.
If obj's class has an image file, use the image from that.
Otherwise create a canvas text item.
Store the icon in the object_icon dictionary and re-use
as needed."""
cell = obj.location
xy = self.cell_topleft(cell)
# Look for an image file
try:
tk_image = self.get_image(obj.get_image_file())
icon = self.create_image(xy, anchor="nw", image=tk_image)
except NoImageException:
# Last resort: create a canvas text icon
icon = self.create_text(xy, anchor="nw", justify="left",
# Abbreviate class name to fit cell
text=obj.__class__.__name__[0:6]
# , fill = ?
#, font = ?
)
# Store and return the icon
self.object_icon[obj] = icon
return icon
def get_image (self, file):
"""Try to find the image in the images dictionary.
If it's not there, open the file and create it, and stick
it in the dictionary. Return the image in a form usable
by the canvas."""
if self.images.has_key(file):
tk_image = self.images[file]
else:
pil_image = Image.open(file)
tk_image = ImageTk.PhotoImage(pil_image)
self.images[file] = tk_image
return tk_image
def user_left(self, event):
print 'left at %d, %d' % self.event_cell(event)
def user_edit_objects(self, event):
"""Choose an object within radius and edit its fields."""
pass
def user_add_object(self, event):
"""Pops up a menu of Object classes; you choose the
one you want to put in this square."""
cell = self.event_cell(event)
xy = self.cell_topleft(cell)
menu = tk.Menu(self, title='Edit (%d, %d)' % cell)
# Generalize object classes available,
# and why is self.run the command?
#for (txt, cmd) in [('Wumpus', self.run), ('Pit', self.run)]:
# menu.add_command(label=txt, command=cmd)
obj_classes = self.env.object_classes()
def class_cmd (oclass):
def cmd ():
obj = oclass()
self.env.add_object(obj, cell)
return cmd
for oclass in obj_classes:
menu.add_command(label=oclass.__name__,
command=class_cmd(oclass))
menu.tk_popup(event.x + self.winfo_rootx(),
event.y + self.winfo_rooty())
def object_added (self, obj):
# Assert obj exists in the environment but has no icon yet
self.add_object_icon(obj)
def object_moved (self, obj):
# Assert obj exists and has an icon already
icon = self.object_icon[obj]
self.coords(icon, self.cell_topleft(obj.location))
def object_deleted (self, obj):
icon = self.object_icon[obj]
del self.object_icon[obj]
self.delete(icon)
def event_cell (self, event):
return self.xy_cell((event.x, event.y))
def xy_cell (self, (x, y)):
"""Given an (x, y) on the canvas, return the row and column
of the cell containing it."""
w = self.cellwidth
return x / w, y / w
def cell_topleft (self, (row, column)):
"""Given a (row, column) tuple, return the (x, y) coordinates
of the cell(row, column)'s top left corner."""
w = self.cellwidth
return w * row, w * column
class NoImageException (Exception): pass
def test_gui ():
v = VacuumEnvironment()
w = EnvGUI(v)
a = TraceAgent(RandomAgent(['Forward',
'TurnRight',
'TurnLeft',
# omit grab because grabbing is
# is not implemented correctly.
#'Grab',
'Suck']))
a.heading = (1, 0) # east?
v.add_object(Dirt(), (6, 6))
v.add_object(Dirt(), (3, 2))
v.add_object(Dirt(), (5, 7))
v.add_object(a, (6, 6))
w.mainloop()
# test_gui()