More on Objects

Introduction

• Previous lecture introduced basics of object-oriented programming
• This one goes a little further
  • As before, the ideas are (almost) universal, but the form varies from language to language
• Send comments

You Can Skip This Lecture If...

• You know what an overloaded operator is
• You know what a static member is
• You know what a design pattern is
• Send comments

Length

• We've already met some special methods, like __init__ and __str__
  • Usually not called directly
  • Instead, Python automatically invokes them at specific times (object creation and string creation)
• There are lots of other special methods
  • For example, if obj has a __len__ method, Python calls it whenever it sees len(obj)

  • class Recent(object):
    
        def __init__(self, number=3):
            self.number = number
            self.items = []
    
        def __str__(self):
            return str(self.items)
    
        def add(self, item):
            self.items.append(item)
            self.items = self.items[-self.number:]
    
        def __len__(self):
            return len(self.items)
    
    if __name__ == '__main__':
        history = Recent()
        for era in ['Permian', 'Trassic', 'Jurassic', 'Cretaceous', 'Tertiary']:
            history.add(era)
            print len(history), history
    

    1 ['Permian']
    2 ['Permian', 'Trassic']
    3 ['Permian', 'Trassic', 'Jurassic']
    3 ['Trassic', 'Jurassic', 'Cretaceous']
    3 ['Jurassic', 'Cretaceous', 'Tertiary']
    

• Send comments

Overloading Operators

• The expression "a + b" is “just” a shorthand for add(a, b)
• Or, if a is an object, for a.add(b)
• But since people might actually want to use the name add, Python spells this method __add__
  • Defining it is an example of operator overloading
• If a class defines a method __add__, it is called whenever something is +'d to the object
  • I.e., x + y calls x.__add__(y)

  • class Recent(object):
        def __add__(self, item):
            self.items.append(item)
            self.items = self.items[-self.number:]
            return self
    if __name__ == '__main__':
        history = Recent()
        for era in ['Permian', 'Trassic', 'Jurassic', 'Cretaceous', 'Tertiary']:
            history = history + era
            print len(history), history

    1 ['Permian']
    2 ['Permian', 'Trassic']
    3 ['Permian', 'Trassic', 'Jurassic']
    3 ['Trassic', 'Jurassic', 'Cretaceous']
    3 ['Jurassic', 'Cretaceous', 'Tertiary']
    

• Send comments

Commutativity

• 2 + x and x + 2 don't always do the same thing
  • Matrix multiplication
• Classes can define right-hand versions of operators, e.g., __radd__ instead of __add__
  • If the object on the left has an __add__ method, call that
  • Otherwise, if the object on the right has an __radd__ method, call that
  • Otherwise, try Python's built-ins
• Send comments

Other Special Methods

• (Almost) every aspect of an object's behavior can be overridden by defining the right method(s)

Method	Purpose
__lt__(self, other)	Less than comparison; __le__, __ne__, and others are used for less than or equal, not equal, etc.
__call__(self, args…)	Called for obj(3, "lithium")
__len__(self)	Object “length”
__getitem__(self, key)	Called for obj[3.14]
__setitem__(self, key, value)	Called for obj[3.14] = 2.17
__contains__	Called for "lithium" in obj
__add__	Called for obj + value; use __mul__ for obj * value, etc.
__int__	Called for int(obj); use __float__ and others to convert to other types
Table 15.1: Special Methods

• Send comments

Example: Sparse Vector

• A vector is sparse if most of its entries are zero
  • Use a dictionary to record non-zero values and their indices
  • No point padding eleven actual values with nine million zeroes
• Overload operators to make the object look like a “real” vector:
  • Addition: create a new vector with a non-zero value wherever either operand had a non-zero value
  • Dot product: add up products of matching non-zero values
  • Length: return one more than the index of the largest non-zero value
    • “One more” to be consistent with Python's lists
• Send comments

How Long is a Sparse Vector?

• What is the length of v after the following operations?

  v = SparseVector() # all values initialized to 0.0
  v[27] = 1.0        # length is now 28
  v[43] = 1.0        # length is now 44
  v[43] = 0.0        # is the length still 44, or 28?
  

• This isn't really a programming question
  • “Largest current index” and “largest index ever seen” can both be implemented
  • The latter is easy, so we'll use that
• Send comments

Vector Behavior

• Construction creates an empty sparse vector
• Define __len__, __getitem__, and __setitem__ to make it behave like a list
  • Exercise: implement del sparse[index]

class SparseVector(object):
    '''Implement a sparse vector.  If a value has not been set
    explicitly, its value is zero.'''

    def __init__(self):
        '''Construct a sparse vector with all zero entries.'''
        self.data = {}

    def __len__(self):
        '''The length of a vector is one more than the largest index.'''
        if self.data:
            return 1 + max(self.data.keys())
        return 0

    def __getitem__(self, key):
        '''Return an explicit value, or 0.0 if none has been set.'''
        if key in self.data:
            return self.data[key]
        return 0.0

    def __setitem__(self, key, value):
        '''Assign a new value to a vector entry.'''
        if type(key) is not int:
            raise KeyError, 'non-integer index to sparse vector'
        self.data[key] = value

• Send comments

Dot Product

• The other object (on the right side of "*") is usually called other
  • No reason to insist that it be a sparse vector
  • Could equally well be a list of values
• So loop over our indices, and multiply by corresponding values in other object
  • Any index not encountered in this loop doesn't matter, since it corresponds to something that's zero
• And make __rmul__ = __mul__ do the same thing as __rmul__

    def __mul__(self, other):
        '''Calculate dot product of a sparse vector with something else.'''

        result = 0.0
        for k in self.data:
            result += self.data[k] * other[k]
        return result

    def __rmul__(self, other):
        return self.__mul__(other)

• Send comments

Addition

• Trickier than multiplication: result is non-zero wherever either argument is non-zero
• Don't want to loop over all the zeroes of either argument
• Solution: if the other object is a sparse vector, cheat
  • I.e., reach inside it, and rely on details of its implementation

    def __add__(self, other):
        '''Add something to a sparse vector.'''

        # Initialize result with all non-zero values from this vector.
        result = SparseVector()
        result.data.update(self.data)

        # If the other object is also a sparse vector, add non-zero values.
        if isinstance(other, SparseVector):
            for k in other.data:
                result[k] = result[k] + other[k]

        # Otherwise, use brute force.
        else:
            for i in range(len(other)):
                result[i] = result[i] + other[i]

        return result

    # Right-hand add does the same thing as left-hand add.
    __radd__ = __add__

• Send comments

Testing

• The class isn't written until the tests are finished
  • Exercise: replace the print statements with assertions

if __name__ == '__main__':

    x = SparseVector()
    x[1] = 1.0
    x[3] = 3.0
    x[5] = 5.0
    print 'len(x)', len(x)
    for i in range(len(x)):
        print '...', i, x[i]

    y = SparseVector()
    y[1] = 10.0
    y[2] = 20.0
    y[3] = 30.0

    print 'x + y', x + y
    print 'y + x', y + x

    print 'x * y', x * y
    print 'y * x', y * x

    z = [0.0, 0.1, 0.2, 0.3, 0.4, 0.5]

    print 'x + z', x + z
    print 'x * z', x * z
    print 'z + x', z + x

len(x) 6
... 0 0.0
... 1 1.0
... 2 0.0
... 3 3.0
... 4 0.0
... 5 5.0
x + y [0.0, 11.0, 20.0, 33.0, 0.0, 5.0]
y + x [0.0, 11.0, 20.0, 33.0, 0.0, 5.0]
x * y 100.0
y * x 100.0
x + z [0.0, 1.1, 0.2, 3.3, 0.4, 5.5]
x * z 3.5
z + x [0.0, 1.1, 0.2, 3.3, 0.4, 5.5]

• Send comments

Static Data Members

• Sometimes want to share data between all instances of a class
  • Constants, a count of the number of class instances created, etc.
• Any data members defined inside the class block belong to the class as a whole

  • class Counter(object):
    
        num = 0     # Number of Counter objects created.
    
        def __init__(self, name):
            Counter.num += 1
            self.name = name
    
    if __name__ == '__main__':
        print 'initial count', Counter.num
        first = Counter('first')
        print 'after creating first object', Counter.num
        second = Counter('second')
        print 'after creating second object', Counter.num
    

    initial count 0
    after creating first object 1
    after creating second object 2
    

• Send comments

Static Methods

• Can also create static methods
  • Just like a function, but put inside the class definition for clarity
• Define the method without the self parameter
  • Since it isn't tied to any particular instance of the class
• Put @staticmethod in front of it
  • A decorator
  • Powerful, but beyond the scope of this course

class Experiment(object):

    already_done = {}

    @staticmethod
    def get_results(name, *params):
        if name in Experiment.already_done:
            return Experiment.already_done[name]
        exp = Experiment(name, *params)
        exp.run()
        Experiment.already_done[name] = exp
        return exp

    def __init__(self, name, *params):
        self.name = name
        self.params = params

    def run(self):
        # marker:vdots
if __name__ == '__main__':
    first = Experiment.get_results('anti-gravity')
    second = Experiment.get_results('time travel')
    third = Experiment.get_results('anti-gravity')
    print 'first ', id(first)
    print 'second', id(second)
    print 'third ', id(third)

first  5120204
second 5120396
third  5120204

• Send comments

Design Patterns

• Style guidelines describe what code should look like line by line
• Design patterns are how we describe larger patterns
  • A standard solution to a commonly-occurring problem
  • That isn't specific enough to be captured once and for all in a library routine or framework
• Idea developed from the 1960s on by the (building) architect Christopher Alexander
  • For example, it's hard to define what a porch is, but the basic idea comes up everywhere the climate is warm
• Introduced to programmers in [Gamma et al 1995]
  • Still a bestseller, but not particularly approachable
• Send comments

The Singleton Pattern

• Problem: want to ensure that there's only ever one instance of a particular class
  • E.g. the controller for a radio telescope antenna
• Considerations:
  • There must be exactly one instance of the class
  • All objects that use the class must have access to that instance
• Solution:
  • Create objects by calling a function instead of the class's constructor
  • Have the function store a reference to the first object it creates
  • Have it return that same object on every subsequent call
• Send comments

Singleton Implementation

class AntennaClass(object):
    '''Singleton that controls a radio telescope.'''

    # The unique instance of the class.
    instance = None

    # The constructor fails if an instance already exists.
    def __init__(self, max_rotation):
        assert AntennaClass.instance is None, 'Trying to create a second instance!'
        self.max_rotation = max_rotation
        AntennaClass.instance = self

# Make the creation function look like a class constructor.
def Antenna(max_rotation):
    '''Create and store an AntennaClass instance, or return the one
    that has already been created.'''
    if AntennaClass.instance:
        return AntennaClass.instance
    return AntennaClass(max_rotation)

• Send comments

Demonstration

first = Antenna(23.5)
print 'first instance:', id(first)
second = Antenna(47.25)
print 'second instance:', id(second)

first instance: 10685200
second instance: 10685200

• Send comments

The Visitor Pattern

• Problem: want an easy way to walk around a complex structure
  • E.g. visit each value in a list of lists of lists exactly once
• Considerations:
  • Many different operations may need to be performed
  • Structure is complex enough that visiting elements is error-prone
  • The types of objects in the structure, and the ways they are connected, are fixed
• Solution:
  • Create a class that knows how to get to each value in turn
  • Give it an empty method that is called once for each value
  • Users derive from this class and fill in the method
  • An all-in-one version of the framework shown earlier
• Send comments

Visitor Implementation

class NestedListVisitor(object):
    '''Visit each element in a list of nested lists.'''

    def __init__(self, data):
        '''Construct, but do not run.'''
        assert type(data) is list, 'Only works on lists!'
        self.data = data

    def run(self):
        '''Iterate over all values.'''
        self.recurse(self.data)

    def recurse(self, current):
        '''Loop over a particular list or sub-list (not meant
        to be called by users).'''
        if type(current) is list:
            for v in current:
                self.recurse(v)
        else:
            self.visit(current)

    def visit(self, value):
        '''Users should fill this method in.'''
        pass

• Send comments

Demonstration

class MaxOfN(NestedListVisitor):

    def __init__(self, data):
        NestedListVisitor.__init__(self, data)
        self.max = None
        self.count = 0

    def visit(self, value):
        self.count += 1
        if self.max is None:
            self.max = value
        else:
            self.max = max(self.max, value)

test_data = [['gold', 'lead'], 'zinc', [['silver', 'iron'], 'mercury']]
test = MaxOfN(test_data)
test.run()
print 'max:', test.max
print 'count:', test.count

max: zinc
count: 6

• Send comments

The Abstract Factory Pattern

• Problem: application doesn't know the specific types of objects it wants to create until runtime
  • If the chromatograph is an RCT-100, create an RCT-100 controller and an RCT-100 configuration panel
  • If it's a Subalta 4C, create a Subalta 4C controller and configuration panel
• Considerations:
  • Objects can be grouped by category and family
    • 

      Figure .:

  • New categories or families may appear later
• Solution:
  • Create a class that knows how to build an instance of each category for a particular family
  • Create another class that stores instances of these builder classes, and calls their methods when asked to
    • Adding a new family is easy, but adding a new category requires changes to every builder
• Send comments

Abstract Factory Builder

class AbstractFamily(object):
    '''Builders for particular families derive from this.'''

    def __init__(self, family):
        self.family = family

    def get_name(self):
        return self.name

    def make_controller(self):
        raise NotImplementedError('make_controller missing')

    def make_configuration_panel(self):
        raise NotImplementedError('make_configuration_panel missing')

• Send comments

Abstract Factory Manager

class FactoryManager(object):
    '''Manage builders by family.'''

    def __init__(self, current_family=None):
        self.builders = {}
        self.family = family

    def set_family(self, family):
        assert family, 'Empty family'
        self.family = family

    def add(self, builder):
        name = builder.get_name()
        self.builders[name] = builder

    def make_controller(self):
        self._check_state()
        return self.builders[self.family].make_controller()

    def make_configuration_panel(self):
        self._check_state()
        return self.builders[self.family].make_configuration_panel()

    def _check_state(self):
        assert self.family, 'No family specified'
        assert self.family in self.builders, 'Unknown family:', self.family

• Send comments

Demonstration

factory = FactoryManager()
factory.add(RCT100Factory())
factory.add(Subalta4CFactory())
factory.set_family('Subalta4C')
controller = factory.make_controller()
configuration_panel = factory.make_configuration_panel()

• Send comments

The Command Pattern

• Problem: want to be able to control the operation of a complex object
  • Turn on the robot arm, move it, lower it, move it again, etc.
• Considerations:
  • Do not want to have to write an entirely new program for each sequence of operations
  • Want to be able to add new operations
  • Would like to be able to undo operations
• Solution:
  • Create one class for each distinct operation
    • Give the class do, undo, and redo methods
  • Create instances of these classes to represent particular commands
  • Create lists of these instances to control the robot arm
• Send comments

Base Command Class

class AbstractCommand(object):
    '''Base class for commands.'''
    def is_undoable(self):
        return False # by default, can't undo/redo operations
    def do(self, robot):
        raise NotImplementedError("Don't know how to do %s" % self.name)
    def undo(self, robot):
        pass
    def redo(self, robot):
        pass

• Send comments

A Particular Command

class MoveCommand(AbstractCommand):
    '''Move the robot arm.'''

    def __init__(self, x, y, z):
        self.x = x
        self.y = y
        self.z = z

    def is_undoable(self):
        return True

    def do(self, robot):
        robot.translate(self.x, self.y, self.z)

    def undo(self, robot):
        robot.translate(-self.x, -self.y, -self.z)

    def redo(self, robot):
        self.do(robot)

• Send comments

Demonstration

robot = Robot()
commands = [MoveCommand(5.0, 2.0, 2.3),
            RotateCommand(-90.0, 0.0, 0.0),
            MoveCommand(1.0, 2.0, 2.0),
            CloseHandCommand()]
for c in commands:
    c.do(robot)

• Send comments

A Few Others

• Cache: store temporary copies of objects locally to improve performance
• State: record state of program as object so that it can be re-started
• Null Object: use an object that does nothing in place of null
  • Saves testing that object isn't null before doing operations
• Adapter: wrap one object in another to give the first a different interface
  • Usually used to give a new library an interface that's compatible with an old one
• Proxy: use one object as an interface to another
  • Typically, the proxy is local, and the real object is on another machine
• Send comments

Summary

• Overloading, design patterns, and other advanced concepts serve two purposes:
  • Communication: a concise way for designers to communicate with each other
  • Education: gives them a way to communicate what they know to newcomers
• Don't expect to connect them all to your own experience the first time
  • But keep them in mind as you look at new problems
• Send comments