pacman.dcop package

Submodules

pacman.dcop.objects module

class pacman.dcop.objects.AgentDef(name, default_route=1, routes=None, default_hosting_cost=0, hosting_costs=None, **kwargs)

Bases: SimpleRepr

Definition of an agent.

AgentDef objects are used when only the definition of the agent is needed, and not the actual running agents. This is for example the case when computing the computations’ distribution, or when instanciating concrete agents.

Route cost default to 1 because they are typically used as a multiplier for message cost when calculating communication cost. On the other hand, hosting cost default to 0 because they are used in a sum. In order to allow using problem-specific attribute on agents, any named argument passed when creating an AgentDef is available as an attribute

>>> a1 = AgentDef('a1', foo='bar')
>>> a1.name
'a1'
>>> a1.foo
'bar'

name: str

the name of the agent

default_route: float

the default cost of a route when not specified in routes.

routes: dictionary of agents name, as string, to float

attribute a specific route cost between this agent and the agents whose names are used as key in the dictionary

default_hosting_cost

the default hosting for a computation when not specified in hosting_costs.

hosting_costs: dictionary of computation name, as string, to float

attribute a specific cost for hosting the computations whose names are used as key in the dictionary.

kwargs: dictionary string -> any

any extra attribute that should be available on this AgentDef object.

__init__(name, default_route=1, routes=None, default_hosting_cost=0, hosting_costs=None, **kwargs)

Build an AgentDef, only the name is mandatory.

property default_hosting_cost: float

Return type

float

property default_route: float

Return type

float

extra_attr()

Extra attributes for this agent definition.

These extra attributes are the kwargs passed to the constructor. They are typically used to defined extra properties on an agent, like the capacity.

Dictionary of strings to values,

Return type

Dict[str, Any]

hosting_cost(computation)

The cost for hosting a computation.

computation: str

the name of the computation

float

the cost for hosting a computation

>>> agt = AgentDef('a1', default_hosting_cost=3)
>>> agt.hosting_cost('c2')
3
>>> agt.hosting_cost('c3')
3

>>> agt = AgentDef('a1', hosting_costs={'c2': 6})
>>> agt.hosting_cost('c2')
6
>>> agt.hosting_cost('c3')
0

Return type

float

property hosting_costs: Dict[str, float]

Return type

Dict[str, float]

property name: str

Return type

str

route(other_agt)

The route cost between this agent and other_agent.

other_agt: str

the name of the other agent

float

the cost of the route

>>> agt = AgentDef('a1', default_route=5)
>>> agt.route('a2')
5
>>> agt.route('a1')
0

>>> agt = AgentDef('a1', routes={'a2':8})
>>> agt.route('a2')
8
>>> agt.route('a3')
1

Return type

float

property routes: Dict[str, float]

Return type

Dict[str, float]

class pacman.dcop.objects.BinaryVariable(name, initial_value=0)

Bases: Variable

__init__(name, initial_value=0)

clone()

class pacman.dcop.objects.Domain(name, domain_type, values)

Bases: Sized, SimpleRepr, Iterable[Any]

A VariableDomain indicates which are the valid values for variables with this domain. It also indicates the type of environment state represented by there variable : ‘luminosity’, humidity’, etc.

A domain object can be used like a list of value as it support basic list-like operations : ‘in’, ‘len’, iterable…

__init__(name, domain_type, values)

Param

name: name of the domain.

Parameters

• domain_type (str) – a string identifying the kind of value in the domain. For example : ‘luminosity’, ‘humidity’, …

• values (Iterable) – an array containing the values allowed for the variables with this domain.

index(val)

Find the position of a value in the domain

val:

a value to look for in the domain

the index of this value in the domain.

>>> d = Domain('d', 'd', [1, 2, 3])
>>> d.index(2)
1

property name: str

Return type

str

to_domain_value(val)

Find a domain value with the same str representation

This is useful when reading value from a file.

valstr

a string that should match a value in the domain (which may contains non-string values, eg int)

a pair (index, value) where index is the position of the value in the domain and value the actual value that matches val.

>>> d = Domain('d', 'd', [1, 2, 3])
>>> d.to_domain_value('2')
(1, 2)

property type: str

Return type

str

property values: Iterable

Return type

Iterable

class pacman.dcop.objects.ExternalVariable(name, domain, value=None)

Bases: Variable

An external is a variable that is not subject to optimization: its value cannot be changed by DCOP algorithms, which only use it as an input, read-only, parameter. The value of an external variable can still change for external reasons, in that case computation(s) should adapt to the change when appropriate. One can be notified of such change by subscribing to the ExternalVariable.

External variable can be used to represent the value from a sensor for example. : it can actually be changed to match the value read from a real sensor or manually by the user (when using a simulator).

__init__(name, domain, value=None)

clone()

subscribe(callback)

unsubscribe(callback)

property value

class pacman.dcop.objects.Variable(name, domain, initial_value=None)

Bases: SimpleRepr

A DCOP variable.

This class represents the definition of a variable : a name, a domain where the variable can take it’s value and an optional initial value. It is not used to keep track of the current value assigned to the variable.

name: str

Name of the variable. You must use a valid python identifier if you want to use python expression (given as string) to define constraints using this variable.

domain: Domain or Iterable

The domain where this variable can take its value. If an iterable is given a Domain object is automatically created (named after the variable name: d_<var_name>.

initial_value: Any

The initial value assigned to the variable.

__init__(name, domain, initial_value=None)

clone()

cost_for_val(val)

Return type

float

property domain: Domain

Return type

Domain

has_cost = False

property initial_value

property name: str

Return type

str

pacman.dcop.objects.VariableDomain

alias of Domain

class pacman.dcop.objects.VariableNoisyCostFunc(name, domain, cost_func, initial_value=None, noise_level=0.02)

Bases: VariableWithCostFunc

__init__(name, domain, cost_func, initial_value=None, noise_level=0.02)

Parameters

cost_func – a function that returns a cost for each value in the

domain.

clone()

cost_for_val(val)

Return type

float

has_cost = True

property noise_level: float

Return type

float

class pacman.dcop.objects.VariableWithCostDict(name, domain, costs, initial_value=None)

Bases: Variable

__init__(name, domain, costs, initial_value=None)

Parameters

• name (str) – The name of the variable

• domain (Union[Domain, Iterable[Any]]) – A VariableDomain object of a list

• costs (Dict[Any, float]) – a dict that associates a cost for each value in domain

• initial_value – optional, if given must be in the domain

clone()

cost_for_val(val)

Return type

float

has_cost = True

class pacman.dcop.objects.VariableWithCostFunc(name, domain, cost_func, initial_value=None)

Bases: Variable

__init__(name, domain, cost_func, initial_value=None)

Parameters

• name (str) – The name of the variable

• domain (Union[Domain, Iterable[Any]]) – A VariableDomain object of a list

• cost_func (Union[Callable[..., float], ExpressionFunction]) – a function that returns a cost for each value in the

domain. :type initial_value: Optional[Any] :param initial_value: optional, if given must be in the domain

clone()

cost_for_val(val)

Return type

float

has_cost = True

pacman.dcop.objects.create_agents(name_prefix, indexes, default_route=1, routes=None, default_hosting_costs=0, hosting_costs=None, separator='_', **kwargs)

Mass creation of agents definitions.

name_prefix: str

Used as prefix when naming the agents.

indexes: non-tuple iterable of indexes or tuple of iterable of indexes

If it not a tuple, an AgentDef is be created for each of the index. If it is a tuple of iterable, an AgentDef is created for every possible combinations of values from indexes.

default_route: float

The default cost of a route when not specified in routes.

routes: dictionary of agents name, as string, to float

Attribute a specific route cost between this agent and the agents whose names are used as key in the dictionary

default_hosting_costs

The default hosting for a computation when not specified in hosting_costs.

hosting_costs: dictionary of computation name, as string, to float

Attribute a specific cost for hosting the computations whose names are used as key in the dictionary.

separator: str kwargs: dictionary

dict

A dictionary ( index -> AgentDef) where index is a string or a tuple of string.

create_variables

When passing an iterable of indexes: >>> agts = create_agents(‘a’, [‘1’, ‘2’, ‘3’], … default_route=2, default_hosting_costs=7) >>> assert isinstance(agts[‘a2’], AgentDef)

When passing a range: >>> agts = create_agents(‘a’, range(20), … default_route=2, default_hosting_costs=7) >>> assert isinstance(agts[‘a08’], AgentDef)

Return type

Dict[Union[str, Tuple[str, ...]], AgentDef]

pacman.dcop.objects.create_binary_variables(name_prefix, indexes, separator='_')

Mass creation of binary variables.

name_prefix: str

Used as prefix when naming the binary variables.

indexes: non-tuple iterable of indexes or tuple of iterables of indexes

If it not a tuple, a binary variable is be created for each of the index. If it is a tuple of iterable, a binary variable is created for every possible combinations of values from indexes.

separator: str

dict

A dictionary ( index -> Binary variable) where index is a string or a tuple of string.

create_variables

When passing an iterable of indexes: >>> vrs = create_binary_variables(’x_’, [‘a1’, ‘a2’, ‘a3’]) >>> assert isinstance(vrs[‘x_a2’], BinaryVariable)

When passing a tuple of iterables of indexes: >>> vrs = create_binary_variables(’m_’, … ([‘x1’, ‘x2’], … [‘a1’, ‘a2’, ‘a3’])) >>> assert isinstance(vrs[(‘x2’, ‘a3’)], BinaryVariable) >>> assert vrs[(‘x2’, ‘a3’)].name == ‘m_x2_a3’ >>> vrs = create_binary_variables(’m_’, … ([‘x1’, ‘x2’], … [‘a1’, ‘a2’, ‘a3’]), … separator=’B’) >>> assert vrs[(‘x2’, ‘a3’)].name == ‘m_x2Ba3’

Return type

Dict[Union[str, Tuple], BinaryVariable]

pacman.dcop.objects.create_variables(name_prefix, indexes, domain, separator='_')

Mass creation of variables.

name_prefix: str

Used as prefix when naming the variables.

indexes: non-tuple iterable of indexes or tuple of iterables of indexes

If it not a tuple, a variable is be created for each of the index. The index might be a range(see examples).

If it is a tuple of iterable, a variable is created

for every possible combinations of values from indexes.

domain: Domain

The domain for the variables.

separator: str

dict

A dictionary ( index -> variable) where index is a string or a tuple of string.

create_binary_variables

When passing an iterable of indexes: >>> vrs = create_variables(’x_’, [‘a1’, ‘a2’, ‘a3’], … Domain(‘color’, ‘’, [‘R’, ‘G’, ‘B’])) >>> assert isinstance(vrs[‘x_a2’], Variable) >>> assert ‘B’ in vrs[‘x_a3’].domain

When passing a range: >>> vrs = create_variables(‘v’, range(10), … Domain(‘color’, ‘’, [‘R’, ‘G’, ‘B’])) >>> assert isinstance(vrs[‘v2’], Variable) >>> assert ‘B’ in vrs[‘v3’].domain

When passing a tuple of iterables of indexes: >>> vrs = create_variables(’m_’, … ([‘x1’, ‘x2’], … [‘a1’, ‘a2’, ‘a3’]), … Domain(‘color’, ‘’, [‘R’, ‘G’, ‘B’])) >>> assert isinstance(vrs[(‘x2’, ‘a3’)], Variable) >>> assert vrs[(‘x2’, ‘a3’)].name == ‘m_x2_a3’ >>> assert ‘R’ in vrs[(‘x2’, ‘a3’)].domain

Return type

Dict[Union[str, Tuple[str, ...]], Variable]

pacman.dcop.relations module

Module contents