def _rvindexed_subs(self, expr, condition=None):
"""
Substitutes the RandomIndexedSymbol with the RandomSymbol with
same name, distribution and probability as RandomIndexedSymbol.
"""
rvs_expr = random_symbols(expr)
if len(rvs_expr) != 0:
swapdict_expr = {}
for rv in rvs_expr:
if isinstance(rv, RandomIndexedSymbol):
newrv = Bernoulli(rv.name,
p=rv.pspace.process.p,
succ=self.success,
fail=self.failure)
swapdict_expr[rv] = newrv
expr = expr.subs(swapdict_expr)
rvs_cond = random_symbols(condition)
if len(rvs_cond) != 0:
swapdict_cond = {}
if condition is not None:
for rv in rvs_cond:
if isinstance(rv, RandomIndexedSymbol):
newrv = Bernoulli(rv.name,
p=rv.pspace.process.p,
succ=self.success,
fail=self.failure)
swapdict_cond[rv] = newrv
condition = condition.subs(swapdict_cond)
return expr, condition
def set(self):
rvs = [i for i in random_symbols(self.args[1])]
marginalise_out = [i for i in random_symbols(self.args[1]) \
if i not in self.args[1]]
for i in rvs:
if i in marginalise_out:
rvs.remove(i)
return ProductSet((i.pspace.set for i in rvs))
def pdf(self, x):
dist = self.args[0]
randoms = []
for arg in dist.args:
randoms.extend(random_symbols(arg))
if len(randoms) > 1:
raise NotImplementedError("Compound Distributions for more than"
" one random argument is not implemeted yet.")
rand_sym = randoms[0]
if isinstance(dist, SingleFiniteDistribution):
y = Dummy('y', integer=True, negative=False)
_pdf = dist.pmf(y)
else:
y = Dummy('y')
_pdf = dist.pdf(y)
if isinstance(rand_sym.pspace.distribution, SingleFiniteDistribution):
rand_dens = rand_sym.pspace.distribution.pmf(rand_sym)
else:
rand_dens = rand_sym.pspace.distribution.pdf(rand_sym)
rand_sym_dom = rand_sym.pspace.domain.set
if rand_sym.pspace.is_Discrete or rand_sym.pspace.is_Finite:
_pdf = Sum(_pdf*rand_dens, (rand_sym, rand_sym_dom._inf,
rand_sym_dom._sup)).doit()
else:
_pdf = integrate(_pdf*rand_dens, (rand_sym, rand_sym_dom._inf,
rand_sym_dom._sup))
return Lambda(y, _pdf)(x)
def doit(self, **hints):
condition = self.args[0]
given_condition = self._condition
numsamples = hints.get('numsamples', False)
for_rewrite = not hints.get('for_rewrite', False)
if isinstance(condition, Not):
return S.One - self.func(condition.args[0],
given_condition,
evaluate=for_rewrite).doit(**hints)
if condition.has(RandomIndexedSymbol):
return pspace(condition).probability(condition,
given_condition,
evaluate=for_rewrite)
if isinstance(given_condition, RandomSymbol):
condrv = random_symbols(condition)
if len(condrv) == 1 and condrv[0] == given_condition:
from sympy.stats.frv_types import BernoulliDistribution
return BernoulliDistribution(
self.func(condition).doit(**hints), 0, 1)
if any([dependent(rv, given_condition) for rv in condrv]):
return Probability(condition, given_condition)
else:
return Probability(condition).doit()
if given_condition is not None and \
not isinstance(given_condition, (Relational, Boolean)):
raise ValueError(
"%s is not a relational or combination of relationals" %
(given_condition))
if given_condition == False or condition is S.false:
return S.Zero
if not isinstance(condition, (Relational, Boolean)):
raise ValueError(
"%s is not a relational or combination of relationals" %
(condition))
if condition is S.true:
return S.One
if numsamples:
return sampling_P(condition,
given_condition,
numsamples=numsamples)
if given_condition is not None: # If there is a condition
# Recompute on new conditional expr
return Probability(given(condition, given_condition)).doit()
# Otherwise pass work off to the ProbabilitySpace
if pspace(condition) == PSpace():
return Probability(condition, given_condition)
result = pspace(condition).probability(condition)
if hasattr(result, 'doit') and for_rewrite:
return result.doit()
else:
return result
def where(self, condition):
rvs = frozenset(random_symbols(condition))
if not (len(rvs) == 1 and rvs.issubset(self.values)):
raise NotImplementedError("Multiple continuous random variables not supported")
rv = tuple(rvs)[0]
interval = reduce_rational_inequalities_wrap(condition, rv)
interval = interval.intersect(self.domain.set)
return SingleContinuousDomain(rv.symbol, interval)
def probability(self,
condition,
given_condition=None,
evaluate=True,
**kwargs):
"""
Computes probability.
Parameters
==========
condition: Relational
Condition for which probability has to be computed. Must
contain a RandomIndexedSymbol of the process.
given_condition: Relational/And
The given conditions under which computations should be done.
Returns
=======
Probability of the condition.
"""
new_condition, new_givencondition = self._rvindexed_subs(
condition, given_condition)
if isinstance(new_givencondition, RandomSymbol):
condrv = random_symbols(new_condition)
if len(condrv) == 1 and condrv[0] == new_givencondition:
return BernoulliDistribution(self.probability(new_condition),
0, 1)
if any([dependent(rv, new_givencondition) for rv in condrv]):
return Probability(new_condition, new_givencondition)
else:
return self.probability(new_condition)
if new_givencondition is not None and \
not isinstance(new_givencondition, (Relational, Boolean)):
raise ValueError(
"%s is not a relational or combination of relationals" %
(new_givencondition))
if new_givencondition == False:
return S.Zero
if new_condition == True:
return S.One
if new_condition == False:
return S.Zero
if not isinstance(new_condition, (Relational, Boolean)):
raise ValueError(
"%s is not a relational or combination of relationals" %
(new_condition))
if new_givencondition is not None: # If there is a condition
# Recompute on new conditional expr
return self.probability(
given(new_condition, new_givencondition, **kwargs), **kwargs)
return pspace(new_condition).probability(new_condition, **kwargs)
def __new__(cls,dist, rvs):
if not all([isinstance(rv, (Indexed, RandomSymbol))] for rv in rvs):
raise ValueError(filldedent('''Marginal distribution can be
intitialised only in terms of random variables or indexed random
variables'''))
rvs = Tuple.fromiter(rv for rv in rvs)
if not isinstance(dist, JointDistribution) and len(random_symbols(dist)) == 0:
return dist
return Basic.__new__(cls, dist, rvs)
def where(self, condition):
rvs = frozenset(random_symbols(condition))
if not (len(rvs) == 1 and rvs.issubset(self.values)):
raise NotImplementedError(
"Multiple continuous random variables not supported")
rv = tuple(rvs)[0]
interval = reduce_poly_inequalities_wrap(condition, rv)
interval = interval.intersect(self.domain.set)
return SingleContinuousDomain(rv.symbol, interval)
def where(self, condition):
rvs = random_symbols(condition)
assert all(r.symbol in self.symbols for r in rvs)
if (len(rvs) > 1):
raise NotImplementedError(filldedent('''Multivariate discrete
random variables are not yet supported.'''))
conditional_domain = reduce_rational_inequalities_wrap(condition,
rvs[0])
conditional_domain = conditional_domain.intersect(self.domain.set)
return SingleDiscreteDomain(rvs[0].symbol, conditional_domain)
def _compound_check(self, dist):
"""
Checks if the given distribution contains random parameters.
"""
randoms = []
for arg in dist.args:
randoms.extend(random_symbols(arg))
if len(randoms) == 0:
return False
return True
def restricted_domain(self, condition):
rvs = random_symbols(condition)
assert all(r.symbol in self.symbols for r in rvs)
if (len(rvs) > 1):
raise NotImplementedError(filldedent('''Multivariate discrete
random variables are not yet supported.'''))
conditional_domain = reduce_rational_inequalities_wrap(condition,
rvs[0])
conditional_domain = conditional_domain.intersect(self.domain.set)
return conditional_domain
def compute_density(self, expr):
if self._is_symbolic:
rv = list(random_symbols(expr))[0]
k = Dummy('k', integer=True)
cond = True if not isinstance(expr, (Relational, Logic)) \
else expr.subs(rv, k)
return Lambda(k,
Piecewise((self.pmf(k), And(k >= self.args[1].low,
k <= self.args[1].high, cond)), (0, True)))
expr = rv_subs(expr, self.values)
return FinitePSpace(self.domain, self.distribution).compute_density(expr)
def pdf(self, *x):
expr, rvs = self.args[0], self.args[1]
marginalise_out = [i for i in random_symbols(expr) if i not in rvs]
if isinstance(expr, JointDistribution):
count = len(expr.domain.args)
x = Dummy('x', real=True, finite=True)
syms = tuple(Indexed(x, i) for i in count)
expr = expr.pdf(syms)
else:
syms = tuple(rv.pspace.symbol if isinstance(rv, RandomSymbol) else rv.args[0] for rv in rvs)
return Lambda(syms, self.compute_pdf(expr, marginalise_out))(*x)
def compute_expectation(self, expr, rvs=None, **kwargs):
if self._is_symbolic:
rv = random_symbols(expr)[0]
k = Dummy('k', integer=True)
expr = expr.subs(rv, k)
cond = True if not isinstance(expr, (Relational, Logic)) \
else expr
func = self.pmf(k) * k if cond != True else self.pmf(k) * expr
return Sum(Piecewise((func, cond), (0, True)),
(k, self.distribution.low, self.distribution.high)).doit()
expr = rv_subs(expr, rvs)
return FinitePSpace(self.domain, self.distribution).compute_expectation(expr, rvs, **kwargs)
def probability(self, condition):
cond_symbols = frozenset(rs.symbol for rs in random_symbols(condition))
cond = rv_subs(condition)
if not cond_symbols.issubset(self.symbols):
raise ValueError("Cannot compare foreign random symbols, %s"
%(str(cond_symbols - self.symbols)))
if isinstance(condition, Relational) and \
(not cond.free_symbols.issubset(self.domain.free_symbols)):
rv = condition.lhs if isinstance(condition.rhs, Symbol) else condition.rhs
return sum(Piecewise(
(self.prob_of(elem), condition.subs(rv, list(elem)[0][1])),
(0, True)) for elem in self.domain)
return sum(self.prob_of(elem) for elem in self.where(condition))
def JointRV(symbol, pdf, _set=None):
"""
Create a Joint Random Variable where each of its component is continuous,
given the following:
Parameters
==========
symbol : Symbol
Represents name of the random variable.
pdf : A PDF in terms of indexed symbols of the symbol given
as the first argument
NOTE
====
As of now, the set for each component for a ``JointRV`` is
equal to the set of all integers, which cannot be changed.
Examples
========
>>> from sympy import exp, pi, Indexed, S
>>> from sympy.stats import density, JointRV
>>> x1, x2 = (Indexed('x', i) for i in (1, 2))
>>> pdf = exp(-x1**2/2 + x1 - x2**2/2 - S(1)/2)/(2*pi)
>>> N1 = JointRV('x', pdf) #Multivariate Normal distribution
>>> density(N1)(1, 2)
exp(-2)/(2*pi)
Returns
=======
RandomSymbol
"""
#TODO: Add support for sets provided by the user
symbol = sympify(symbol)
syms = list(i for i in pdf.free_symbols if isinstance(i, Indexed)
and i.base == IndexedBase(symbol))
syms = tuple(sorted(syms, key = lambda index: index.args[1]))
_set = S.Reals**len(syms)
pdf = Lambda(syms, pdf)
dist = JointDistributionHandmade(pdf, _set)
jrv = JointPSpace(symbol, dist).value
rvs = random_symbols(pdf)
if len(rvs) != 0:
dist = MarginalDistribution(dist, (jrv,))
return JointPSpace(symbol, dist).value
return jrv
def doit(self, **hints):
deep = hints.get('deep', True)
condition = self._condition
expr = self.args[0]
numsamples = hints.get('numsamples', False)
for_rewrite = not hints.get('for_rewrite', False)
if deep:
expr = expr.doit(**hints)
if not random_symbols(expr) or isinstance(
expr, Expectation): # expr isn't random?
return expr
if numsamples: # Computing by monte carlo sampling?
evalf = hints.get('evalf', True)
return sampling_E(expr,
condition,
numsamples=numsamples,
evalf=evalf)
if expr.has(RandomIndexedSymbol):
return pspace(expr).compute_expectation(expr, condition)
# Create new expr and recompute E
if condition is not None: # If there is a condition
return self.func(given(expr, condition)).doit(**hints)
# A few known statements for efficiency
if expr.is_Add: # We know that E is Linear
return Add(*[
self.func(arg, condition).doit(
**hints) if not isinstance(arg, Expectation) else self.
func(arg, condition) for arg in expr.args
])
if expr.is_Mul:
if expr.atoms(Expectation):
return expr
if pspace(expr) == PSpace():
return self.func(expr)
# Otherwise case is simple, pass work off to the ProbabilitySpace
result = pspace(expr).compute_expectation(expr, evaluate=for_rewrite)
if hasattr(result, 'doit') and for_rewrite:
return result.doit(**hints)
else:
return result
def pdf(self, *x):
expr, rvs = self.args[0], self.args[1]
marginalise_out = [i for i in random_symbols(expr) if i not in self.args[1]]
syms = [i.pspace.symbol for i in self.args[1]]
for i in expr.atoms(Indexed):
if isinstance(i, Indexed) and isinstance(i.base, RandomSymbol)\
and i not in rvs:
marginalise_out.append(i)
if isinstance(expr, CompoundDistribution):
syms = Dummy('x', real=True)
expr = expr.args[0].pdf(syms)
elif isinstance(expr, JointDistribution):
count = len(expr.domain.args)
x = Dummy('x', real=True, finite=True)
syms = [Indexed(x, i) for i in count]
expr = expression.pdf(syms)
return Lambda(syms, self.compute_pdf(expr, marginalise_out))(*x)
def compute_expectation(self, expr, rvs=None, evaluate=False, **kwargs):
syms = tuple(self.value[i] for i in range(self.component_count))
rvs = rvs or syms
if not any([i in rvs for i in syms]):
return expr
expr = expr*self.pdf
for rv in rvs:
if isinstance(rv, Indexed):
expr = expr.xreplace({rv: Indexed(str(rv.base), rv.args[1])})
elif isinstance(rv, RandomSymbol):
expr = expr.xreplace({rv: rv.symbol})
if self.value in random_symbols(expr):
raise NotImplementedError(filldedent('''
Expectations of expression with unindexed joint random symbols
cannot be calculated yet.'''))
limits = tuple((Indexed(str(rv.base),rv.args[1]),
self.distribution.set.args[rv.args[1]]) for rv in syms)
return Integral(expr, *limits)
def compute_expectation(self, expr, rvs=None, evaluate=False, **kwargs):
syms = tuple(self.value[i] for i in range(self.component_count))
rvs = rvs or syms
if not any([i in rvs for i in syms]):
return expr
expr = expr*self.pdf
for rv in rvs:
if isinstance(rv, Indexed):
expr = expr.xreplace({rv: Indexed(str(rv.base), rv.args[1])})
elif isinstance(rv, RandomSymbol):
expr = expr.xreplace({rv: rv.symbol})
if self.value in random_symbols(expr):
raise NotImplementedError(filldedent('''
Expectations of expression with unindexed joint random symbols
cannot be calculated yet.'''))
limits = tuple((Indexed(str(rv.base),rv.args[1]),
self.distribution.set.args[rv.args[1]]) for rv in syms)
return Integral(expr, *limits)
def JointRV(symbol, pdf, _set=None):
"""
Create a Joint Random Variable where each of its component is conitinuous,
given the following:
-- a symbol
-- a PDF in terms of indexed symbols of the symbol given
as the first argument
NOTE: As of now, the set for each component for a `JointRV` is
equal to the set of all integers, which can not be changed.
Returns a RandomSymbol.
Examples
========
>>> from sympy import symbols, exp, pi, Indexed, S
>>> from sympy.stats import density
>>> from sympy.stats.joint_rv_types import JointRV
>>> x1, x2 = (Indexed('x', i) for i in (1, 2))
>>> pdf = exp(-x1**2/2 + x1 - x2**2/2 - S(1)/2)/(2*pi)
>>> N1 = JointRV('x', pdf) #Multivariate Normal distribution
>>> density(N1)(1, 2)
exp(-2)/(2*pi)
"""
#TODO: Add support for sets provided by the user
symbol = sympify(symbol)
syms = list(i for i in pdf.free_symbols if isinstance(i, Indexed)
and i.base == IndexedBase(symbol))
syms.sort(key = lambda index: index.args[1])
_set = S.Reals**len(syms)
pdf = Lambda(syms, pdf)
dist = JointDistributionHandmade(pdf, _set)
jrv = JointPSpace(symbol, dist).value
rvs = random_symbols(pdf)
if len(rvs) != 0:
dist = MarginalDistribution(dist, (jrv,))
return JointPSpace(symbol, dist).value
return jrv
def probability(self, condition, **kwargs):
z = Dummy('z', real=True)
cond_inv = False
if isinstance(condition, Ne):
condition = Eq(condition.args[0], condition.args[1])
cond_inv = True
# Univariate case can be handled by where
try:
domain = self.where(condition)
rv = [rv for rv in self.values if rv.symbol == domain.symbol][0]
# Integrate out all other random variables
pdf = self.compute_density(rv, **kwargs)
# return S.Zero if `domain` is empty set
if domain.set is S.EmptySet or isinstance(domain.set, FiniteSet):
return S.Zero if not cond_inv else S.One
if isinstance(domain.set, Union):
return sum(
Integral(pdf(z), (z, subset), **kwargs)
for subset in domain.set.args
if isinstance(subset, Interval))
# Integrate out the last variable over the special domain
return Integral(pdf(z), (z, domain.set), **kwargs)
# Other cases can be turned into univariate case
# by computing a density handled by density computation
except NotImplementedError:
from sympy.stats.rv import density
expr = condition.lhs - condition.rhs
if not random_symbols(expr):
dens = self.density
comp = condition.rhs
else:
dens = density(expr, **kwargs)
comp = 0
if not isinstance(dens, ContinuousDistribution):
from sympy.stats.crv_types import ContinuousDistributionHandmade
dens = ContinuousDistributionHandmade(dens,
set=self.domain.set)
# Turn problem into univariate case
space = SingleContinuousPSpace(z, dens)
result = space.probability(condition.__class__(space.value, comp))
return result if not cond_inv else S.One - result
def probability(self, condition):
cond_symbols = frozenset(rs.symbol for rs in random_symbols(condition))
assert cond_symbols.issubset(self.symbols)
return sum(self.prob_of(elem) for elem in self.where(condition))
def where(self, condition):
assert all(r.symbol in self.symbols for r in random_symbols(condition))
return ConditionalFiniteDomain(self.domain, condition)
def where(self, condition):
assert all(r.symbol in self.symbols for r in random_symbols(condition))
return ConditionalFiniteDomain(self.domain, condition)
def domain(self):
rvs = random_symbols(self.distribution)
if not rvs:
return SingleDomain(self.symbol, self.distribution.set)
return ProductDomain(*[rv.pspace.domain for rv in rvs])
def latent_distributions(self):
return random_symbols(self.args[0])
def probability(self, condition):
cond_symbols = frozenset(rs.symbol for rs in random_symbols(condition))
assert cond_symbols.issubset(self.symbols)
return sum(self.prob_of(elem) for elem in self.where(condition))
def expectation(self, expr, condition=None, evaluate=True, **kwargs):
"""
Handles expectation queries for markov process.
Parameters
==========
expr: RandomIndexedSymbol, Relational, Logic
Condition for which expectation has to be computed. Must
contain a RandomIndexedSymbol of the process.
condition: Relational, Logic
The given conditions under which computations should be done.
Returns
=======
Expectation
Unevaluated object if computations cannot be done due to
insufficient information.
Expr
In all other cases when the computations are successful.
Note
====
Any information passed at the time of query overrides
any information passed at the time of object creation like
transition probabilities, state space.
Pass the transition matrix using TransitionMatrixOf,
generator matrix using GeneratorMatrixOf and state space
using StochasticStateSpaceOf in given_condition using & or And.
"""
check, mat, state_space, condition = \
self._preprocess(condition, evaluate)
if check:
return Expectation(expr, condition)
rvs = random_symbols(expr)
if isinstance(expr, Expr) and isinstance(condition, Eq) \
and len(rvs) == 1:
# handle queries similar to E(f(X[i]), Eq(X[i-m], <some-state>))
rv = list(rvs)[0]
lhsg, rhsg = condition.lhs, condition.rhs
if not isinstance(lhsg, RandomIndexedSymbol):
lhsg, rhsg = (rhsg, lhsg)
if rhsg not in self.state_space:
raise ValueError("%s state is not in the state space." %
(rhsg))
if rv.key < lhsg.key:
raise ValueError(
"Incorrect given condition is given, expectation "
"time %s < time %s" % (rv.key, rv.key))
mat_of = TransitionMatrixOf(self, mat) if isinstance(
self, DiscreteMarkovChain) else GeneratorMatrixOf(self, mat)
cond = condition & mat_of & \
StochasticStateSpaceOf(self, state_space)
func = lambda s: self.probability(Eq(rv, s), cond) * expr.subs(
rv, s)
return sum([func(s) for s in state_space])
raise NotImplementedError(
"Mechanism for handling (%s, %s) queries hasn't been "
"implemented yet." % (expr, condition))
def domain(self):
rvs = random_symbols(self.distribution)
if len(rvs) == 0:
return SingleDomain(self.symbol, self.set)
return ProductDomain(*[rv.pspace.domain for rv in rvs])
