You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

486 lines
14 KiB

5 months ago
"""A module which implements predicates and assumption context."""
from contextlib import contextmanager
import inspect
from sympy.core.symbol import Str
from sympy.core.sympify import _sympify
from sympy.logic.boolalg import Boolean, false, true
from sympy.multipledispatch.dispatcher import Dispatcher, str_signature
from sympy.utilities.exceptions import sympy_deprecation_warning
from sympy.utilities.iterables import is_sequence
from sympy.utilities.source import get_class
class AssumptionsContext(set):
"""
Set containing default assumptions which are applied to the ``ask()``
function.
Explanation
===========
This is used to represent global assumptions, but you can also use this
class to create your own local assumptions contexts. It is basically a thin
wrapper to Python's set, so see its documentation for advanced usage.
Examples
========
The default assumption context is ``global_assumptions``, which is initially empty:
>>> from sympy import ask, Q
>>> from sympy.assumptions import global_assumptions
>>> global_assumptions
AssumptionsContext()
You can add default assumptions:
>>> from sympy.abc import x
>>> global_assumptions.add(Q.real(x))
>>> global_assumptions
AssumptionsContext({Q.real(x)})
>>> ask(Q.real(x))
True
And remove them:
>>> global_assumptions.remove(Q.real(x))
>>> print(ask(Q.real(x)))
None
The ``clear()`` method removes every assumption:
>>> global_assumptions.add(Q.positive(x))
>>> global_assumptions
AssumptionsContext({Q.positive(x)})
>>> global_assumptions.clear()
>>> global_assumptions
AssumptionsContext()
See Also
========
assuming
"""
def add(self, *assumptions):
"""Add assumptions."""
for a in assumptions:
super().add(a)
def _sympystr(self, printer):
if not self:
return "%s()" % self.__class__.__name__
return "{}({})".format(self.__class__.__name__, printer._print_set(self))
global_assumptions = AssumptionsContext()
class AppliedPredicate(Boolean):
"""
The class of expressions resulting from applying ``Predicate`` to
the arguments. ``AppliedPredicate`` merely wraps its argument and
remain unevaluated. To evaluate it, use the ``ask()`` function.
Examples
========
>>> from sympy import Q, ask
>>> Q.integer(1)
Q.integer(1)
The ``function`` attribute returns the predicate, and the ``arguments``
attribute returns the tuple of arguments.
>>> type(Q.integer(1))
<class 'sympy.assumptions.assume.AppliedPredicate'>
>>> Q.integer(1).function
Q.integer
>>> Q.integer(1).arguments
(1,)
Applied predicates can be evaluated to a boolean value with ``ask``:
>>> ask(Q.integer(1))
True
"""
__slots__ = ()
def __new__(cls, predicate, *args):
if not isinstance(predicate, Predicate):
raise TypeError("%s is not a Predicate." % predicate)
args = map(_sympify, args)
return super().__new__(cls, predicate, *args)
@property
def arg(self):
"""
Return the expression used by this assumption.
Examples
========
>>> from sympy import Q, Symbol
>>> x = Symbol('x')
>>> a = Q.integer(x + 1)
>>> a.arg
x + 1
"""
# Will be deprecated
args = self._args
if len(args) == 2:
# backwards compatibility
return args[1]
raise TypeError("'arg' property is allowed only for unary predicates.")
@property
def function(self):
"""
Return the predicate.
"""
# Will be changed to self.args[0] after args overriding is removed
return self._args[0]
@property
def arguments(self):
"""
Return the arguments which are applied to the predicate.
"""
# Will be changed to self.args[1:] after args overriding is removed
return self._args[1:]
def _eval_ask(self, assumptions):
return self.function.eval(self.arguments, assumptions)
@property
def binary_symbols(self):
from .ask import Q
if self.function == Q.is_true:
i = self.arguments[0]
if i.is_Boolean or i.is_Symbol:
return i.binary_symbols
if self.function in (Q.eq, Q.ne):
if true in self.arguments or false in self.arguments:
if self.arguments[0].is_Symbol:
return {self.arguments[0]}
elif self.arguments[1].is_Symbol:
return {self.arguments[1]}
return set()
class PredicateMeta(type):
def __new__(cls, clsname, bases, dct):
# If handler is not defined, assign empty dispatcher.
if "handler" not in dct:
name = f"Ask{clsname.capitalize()}Handler"
handler = Dispatcher(name, doc="Handler for key %s" % name)
dct["handler"] = handler
dct["_orig_doc"] = dct.get("__doc__", "")
return super().__new__(cls, clsname, bases, dct)
@property
def __doc__(cls):
handler = cls.handler
doc = cls._orig_doc
if cls is not Predicate and handler is not None:
doc += "Handler\n"
doc += " =======\n\n"
# Append the handler's doc without breaking sphinx documentation.
docs = [" Multiply dispatched method: %s" % handler.name]
if handler.doc:
for line in handler.doc.splitlines():
if not line:
continue
docs.append(" %s" % line)
other = []
for sig in handler.ordering[::-1]:
func = handler.funcs[sig]
if func.__doc__:
s = ' Inputs: <%s>' % str_signature(sig)
lines = []
for line in func.__doc__.splitlines():
lines.append(" %s" % line)
s += "\n".join(lines)
docs.append(s)
else:
other.append(str_signature(sig))
if other:
othersig = " Other signatures:"
for line in other:
othersig += "\n * %s" % line
docs.append(othersig)
doc += '\n\n'.join(docs)
return doc
class Predicate(Boolean, metaclass=PredicateMeta):
"""
Base class for mathematical predicates. It also serves as a
constructor for undefined predicate objects.
Explanation
===========
Predicate is a function that returns a boolean value [1].
Predicate function is object, and it is instance of predicate class.
When a predicate is applied to arguments, ``AppliedPredicate``
instance is returned. This merely wraps the argument and remain
unevaluated. To obtain the truth value of applied predicate, use the
function ``ask``.
Evaluation of predicate is done by multiple dispatching. You can
register new handler to the predicate to support new types.
Every predicate in SymPy can be accessed via the property of ``Q``.
For example, ``Q.even`` returns the predicate which checks if the
argument is even number.
To define a predicate which can be evaluated, you must subclass this
class, make an instance of it, and register it to ``Q``. After then,
dispatch the handler by argument types.
If you directly construct predicate using this class, you will get
``UndefinedPredicate`` which cannot be dispatched. This is useful
when you are building boolean expressions which do not need to be
evaluated.
Examples
========
Applying and evaluating to boolean value:
>>> from sympy import Q, ask
>>> ask(Q.prime(7))
True
You can define a new predicate by subclassing and dispatching. Here,
we define a predicate for sexy primes [2] as an example.
>>> from sympy import Predicate, Integer
>>> class SexyPrimePredicate(Predicate):
... name = "sexyprime"
>>> Q.sexyprime = SexyPrimePredicate()
>>> @Q.sexyprime.register(Integer, Integer)
... def _(int1, int2, assumptions):
... args = sorted([int1, int2])
... if not all(ask(Q.prime(a), assumptions) for a in args):
... return False
... return args[1] - args[0] == 6
>>> ask(Q.sexyprime(5, 11))
True
Direct constructing returns ``UndefinedPredicate``, which can be
applied but cannot be dispatched.
>>> from sympy import Predicate, Integer
>>> Q.P = Predicate("P")
>>> type(Q.P)
<class 'sympy.assumptions.assume.UndefinedPredicate'>
>>> Q.P(1)
Q.P(1)
>>> Q.P.register(Integer)(lambda expr, assump: True)
Traceback (most recent call last):
...
TypeError: <class 'sympy.assumptions.assume.UndefinedPredicate'> cannot be dispatched.
References
==========
.. [1] https://en.wikipedia.org/wiki/Predicate_%28mathematical_logic%29
.. [2] https://en.wikipedia.org/wiki/Sexy_prime
"""
is_Atom = True
def __new__(cls, *args, **kwargs):
if cls is Predicate:
return UndefinedPredicate(*args, **kwargs)
obj = super().__new__(cls, *args)
return obj
@property
def name(self):
# May be overridden
return type(self).__name__
@classmethod
def register(cls, *types, **kwargs):
"""
Register the signature to the handler.
"""
if cls.handler is None:
raise TypeError("%s cannot be dispatched." % type(cls))
return cls.handler.register(*types, **kwargs)
@classmethod
def register_many(cls, *types, **kwargs):
"""
Register multiple signatures to same handler.
"""
def _(func):
for t in types:
if not is_sequence(t):
t = (t,) # for convenience, allow passing `type` to mean `(type,)`
cls.register(*t, **kwargs)(func)
return _
def __call__(self, *args):
return AppliedPredicate(self, *args)
def eval(self, args, assumptions=True):
"""
Evaluate ``self(*args)`` under the given assumptions.
This uses only direct resolution methods, not logical inference.
"""
result = None
try:
result = self.handler(*args, assumptions=assumptions)
except NotImplementedError:
pass
return result
def _eval_refine(self, assumptions):
# When Predicate is no longer Boolean, delete this method
return self
class UndefinedPredicate(Predicate):
"""
Predicate without handler.
Explanation
===========
This predicate is generated by using ``Predicate`` directly for
construction. It does not have a handler, and evaluating this with
arguments is done by SAT solver.
Examples
========
>>> from sympy import Predicate, Q
>>> Q.P = Predicate('P')
>>> Q.P.func
<class 'sympy.assumptions.assume.UndefinedPredicate'>
>>> Q.P.name
Str('P')
"""
handler = None
def __new__(cls, name, handlers=None):
# "handlers" parameter supports old design
if not isinstance(name, Str):
name = Str(name)
obj = super(Boolean, cls).__new__(cls, name)
obj.handlers = handlers or []
return obj
@property
def name(self):
return self.args[0]
def _hashable_content(self):
return (self.name,)
def __getnewargs__(self):
return (self.name,)
def __call__(self, expr):
return AppliedPredicate(self, expr)
def add_handler(self, handler):
sympy_deprecation_warning(
"""
The AskHandler system is deprecated. Predicate.add_handler()
should be replaced with the multipledispatch handler of Predicate.
""",
deprecated_since_version="1.8",
active_deprecations_target='deprecated-askhandler',
)
self.handlers.append(handler)
def remove_handler(self, handler):
sympy_deprecation_warning(
"""
The AskHandler system is deprecated. Predicate.remove_handler()
should be replaced with the multipledispatch handler of Predicate.
""",
deprecated_since_version="1.8",
active_deprecations_target='deprecated-askhandler',
)
self.handlers.remove(handler)
def eval(self, args, assumptions=True):
# Support for deprecated design
# When old design is removed, this will always return None
sympy_deprecation_warning(
"""
The AskHandler system is deprecated. Evaluating UndefinedPredicate
objects should be replaced with the multipledispatch handler of
Predicate.
""",
deprecated_since_version="1.8",
active_deprecations_target='deprecated-askhandler',
stacklevel=5,
)
expr, = args
res, _res = None, None
mro = inspect.getmro(type(expr))
for handler in self.handlers:
cls = get_class(handler)
for subclass in mro:
eval_ = getattr(cls, subclass.__name__, None)
if eval_ is None:
continue
res = eval_(expr, assumptions)
# Do not stop if value returned is None
# Try to check for higher classes
if res is None:
continue
if _res is None:
_res = res
else:
# only check consistency if both resolutors have concluded
if _res != res:
raise ValueError('incompatible resolutors')
break
return res
@contextmanager
def assuming(*assumptions):
"""
Context manager for assumptions.
Examples
========
>>> from sympy import assuming, Q, ask
>>> from sympy.abc import x, y
>>> print(ask(Q.integer(x + y)))
None
>>> with assuming(Q.integer(x), Q.integer(y)):
... print(ask(Q.integer(x + y)))
True
"""
old_global_assumptions = global_assumptions.copy()
global_assumptions.update(assumptions)
try:
yield
finally:
global_assumptions.clear()
global_assumptions.update(old_global_assumptions)