def test_ContinuousMarkovChain(): T1 = Matrix([[S(-2), S(2), S.Zero], [S.Zero, S.NegativeOne, S.One], [Rational(3, 2), Rational(3, 2), S(-3)]]) C1 = ContinuousMarkovChain('C', [0, 1, 2], T1) assert C1.limiting_distribution() == ImmutableMatrix([[Rational(3, 19), Rational(12, 19), Rational(4, 19)]]) T2 = Matrix([[-S.One, S.One, S.Zero], [S.One, -S.One, S.Zero], [S.Zero, S.One, -S.One]]) C2 = ContinuousMarkovChain('C', [0, 1, 2], T2) A, t = C2.generator_matrix, symbols('t', positive=True) assert C2.transition_probabilities(A)(t) == Matrix([[S.Half + exp(-2*t)/2, S.Half - exp(-2*t)/2, 0], [S.Half - exp(-2*t)/2, S.Half + exp(-2*t)/2, 0], [S.Half - exp(-t) + exp(-2*t)/2, S.Half - exp(-2*t)/2, exp(-t)]]) with ignore_warnings(UserWarning): ### TODO: Restore tests once warnings are removed assert P(Eq(C2(1), 1), Eq(C2(0), 1), evaluate=False) == Probability(Eq(C2(1), 1), Eq(C2(0), 1)) assert P(Eq(C2(1), 1), Eq(C2(0), 1)) == exp(-2)/2 + S.Half assert P(Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 1), Eq(P(Eq(C2(1), 0)), S.Half)) == (Rational(1, 4) - exp(-2)/4)*(exp(-2)/2 + S.Half) assert P(Not(Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 2)) | (Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 2)), Eq(P(Eq(C2(1), 0)), Rational(1, 4)) & Eq(P(Eq(C2(1), 1)), Rational(1, 4))) is S.One assert E(C2(Rational(3, 2)), Eq(C2(0), 2)) == -exp(-3)/2 + 2*exp(Rational(-3, 2)) + S.Half assert variance(C2(Rational(3, 2)), Eq(C2(0), 1)) == ((S.Half - exp(-3)/2)**2*(exp(-3)/2 + S.Half) + (Rational(-1, 2) - exp(-3)/2)**2*(S.Half - exp(-3)/2)) raises(KeyError, lambda: P(Eq(C2(1), 0), Eq(P(Eq(C2(1), 1)), S.Half))) assert P(Eq(C2(1), 0), Eq(P(Eq(C2(5), 1)), S.Half)) == Probability(Eq(C2(1), 0)) TS1 = MatrixSymbol('G', 3, 3) CS1 = ContinuousMarkovChain('C', [0, 1, 2], TS1) A = CS1.generator_matrix assert CS1.transition_probabilities(A)(t) == exp(t*A) C3 = ContinuousMarkovChain('C', [Symbol('0'), Symbol('1'), Symbol('2')], T2) assert P(Eq(C3(1), 1), Eq(C3(0), 1)) == exp(-2)/2 + S.Half assert P(Eq(C3(1), Symbol('1')), Eq(C3(0), Symbol('1'))) == exp(-2)/2 + S.Half
def test_ContinuousMarkovChain(): T1 = Matrix([[S(-2), S(2), S(0)], [S(0), S(-1), S(1)], [S(3)/2, S(3)/2, S(-3)]]) C1 = ContinuousMarkovChain('C', [0, 1, 2], T1) assert C1.limiting_distribution() == ImmutableMatrix([[S(3)/19, S(12)/19, S(4)/19]]) T2 = Matrix([[-S(1), S(1), S(0)], [S(1), -S(1), S(0)], [S(0), S(1), -S(1)]]) C2 = ContinuousMarkovChain('C', [0, 1, 2], T2) A, t = C2.generator_matrix, symbols('t', positive=True) assert C2.transition_probabilities(A)(t) == Matrix([[S(1)/2 + exp(-2*t)/2, S(1)/2 - exp(-2*t)/2, 0], [S(1)/2 - exp(-2*t)/2, S(1)/2 + exp(-2*t)/2, 0], [S(1)/2 - exp(-t) + exp(-2*t)/2, S(1)/2 - exp(-2*t)/2, exp(-t)]]) assert P(Eq(C2(1), 1), Eq(C2(0), 1), evaluate=False) == Probability(Eq(C2(1), 1)) assert P(Eq(C2(1), 1), Eq(C2(0), 1)) == exp(-2)/2 + S(1)/2 assert P(Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 1), Eq(P(Eq(C2(1), 0)), S(1)/2)) == (S(1)/4 - exp(-2)/4)*(exp(-2)/2 + S(1)/2) assert P(Not(Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 2)) | (Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 2)), Eq(P(Eq(C2(1), 0)), S(1)/4) & Eq(P(Eq(C2(1), 1)), S(1)/4)) == S(1) assert E(C2(S(3)/2), Eq(C2(0), 2)) == -exp(-3)/2 + 2*exp(-S(3)/2) + S(1)/2 assert variance(C2(S(3)/2), Eq(C2(0), 1)) == ((S(1)/2 - exp(-3)/2)**2*(exp(-3)/2 + S(1)/2) + (-S(1)/2 - exp(-3)/2)**2*(S(1)/2 - exp(-3)/2)) raises(KeyError, lambda: P(Eq(C2(1), 0), Eq(P(Eq(C2(1), 1)), S(1)/2))) assert P(Eq(C2(1), 0), Eq(P(Eq(C2(5), 1)), S(1)/2)) == Probability(Eq(C2(1), 0)) TS1 = MatrixSymbol('G', 3, 3) CS1 = ContinuousMarkovChain('C', [0, 1, 2], TS1) A = CS1.generator_matrix assert CS1.transition_probabilities(A)(t) == exp(t*A)
def test_literal_probability(): X = Normal('X', 2, 3) Y = Normal('Y', 3, 4) Z = Poisson('Z', 4) W = Poisson('W', 3) x, y, w, z = symbols('x, y, w, z') assert Probability(X > 0).doit() == probability(X > 0) assert Probability(X > x).doit() == probability(X > x) assert Expectation(X).doit() == expectation(X) assert Expectation(X**2).doit() == expectation(X**2) assert Expectation(x*X) == x*Expectation(X) assert Expectation(2*X + 3*Y + z*X*Y) == 2*Expectation(X) + 3*Expectation(Y) + z*Expectation(X*Y) assert Expectation(2*X + 3*Y + z*X*Y, evaluate=False).args == (2*X + 3*Y + z*X*Y,) assert Expectation(sin(X)) == Expectation(sin(X), evaluate=False) assert Expectation(2*x*sin(X)*Y + y*X**2 + z*X*Y) == 2*x*Expectation(sin(X)*Y) + y*Expectation(X**2) + z*Expectation(X*Y) assert Variance(w) == 0 assert Variance(X).doit() == variance(X) assert Variance(X + z) == Variance(X) assert Variance(X*Y).args == (Mul(X, Y),) assert type(Variance(X*Y)) == Variance assert Variance(z*X) == z**2*Variance(X) assert Variance(X + Y) == Variance(X) + Variance(Y) + 2*Covariance(X, Y) assert Variance(X + Y + Z + W) == (Variance(X) + Variance(Y) + Variance(Z) + Variance(W) + 2 * Covariance(X, Y) + 2 * Covariance(X, Z) + 2 * Covariance(X, W) + 2 * Covariance(Y, Z) + 2 * Covariance(Y, W) + 2 * Covariance(W, Z)) assert Variance(X**2).doit() == variance(X**2) assert Variance(X**2, evaluate=False) == Variance(X**2) assert Variance(x*X**2) == x**2*Variance(X**2) assert Variance(sin(X)).args == (sin(X),) assert Variance(sin(X), evaluate=False) == Variance(sin(X)) assert Variance(x*sin(X)) == x**2*Variance(sin(X)) assert Covariance(w, z) == 0 assert Covariance(X, w) == 0 assert Covariance(w, X) == 0 assert Covariance(X, Y).args == (X, Y) assert type(Covariance(X, Y)) == Covariance assert Covariance(z*X + 3, Y) == z*Covariance(X, Y) assert Covariance(X, X) == Variance(X) assert Covariance(z*X + 3, w*Y + 4) == w*z*Covariance(X,Y) assert Covariance(X, Y) == Covariance(Y, X) assert Covariance(X + Y, Z + W) == Covariance(W, X) + Covariance(W, Y) + Covariance(X, Z) + Covariance(Y, Z) assert Covariance(x*X + y*Y, z*Z + w*W) == (x*w*Covariance(W, X) + w*y*Covariance(W, Y) + x*z*Covariance(X, Z) + y*z*Covariance(Y, Z)) assert Covariance(x*X**2 + y*sin(Y), z*Y*Z**2 + w*W) == (w*x*Covariance(W, X**2) + w*y*Covariance(sin(Y), W) + x*z*Covariance(Y*Z**2, X**2) + y*z*Covariance(Y*Z**2, sin(Y))) assert Covariance(X, X**2) == Covariance(X, X**2, evaluate=False) assert Covariance(X, sin(X)) == Covariance(sin(X), X, evaluate=False) assert Covariance(X**2, sin(X)*Y) == Covariance(sin(X)*Y, X**2, evaluate=False)
def test_ContinuousMarkovChain(): T1 = Matrix([ [S(-2), S(2), S.Zero], [S.Zero, S.NegativeOne, S.One], [Rational(3, 2), Rational(3, 2), S(-3)], ]) C1 = ContinuousMarkovChain("C", [0, 1, 2], T1) assert C1.limiting_distribution() == ImmutableMatrix( [[Rational(3, 19), Rational(12, 19), Rational(4, 19)]]) T2 = Matrix([[-S.One, S.One, S.Zero], [S.One, -S.One, S.Zero], [S.Zero, S.One, -S.One]]) C2 = ContinuousMarkovChain("C", [0, 1, 2], T2) A, t = C2.generator_matrix, symbols("t", positive=True) assert C2.transition_probabilities(A)(t) == Matrix([ [S.Half + exp(-2 * t) / 2, S.Half - exp(-2 * t) / 2, 0], [S.Half - exp(-2 * t) / 2, S.Half + exp(-2 * t) / 2, 0], [ S.Half - exp(-t) + exp(-2 * t) / 2, S.Half - exp(-2 * t) / 2, exp(-t) ], ]) assert P(Eq(C2(1), 1), Eq(C2(0), 1), evaluate=False) == Probability(Eq(C2(1), 1)) assert P(Eq(C2(1), 1), Eq(C2(0), 1)) == exp(-2) / 2 + S.Half assert P( Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 1), Eq(P(Eq(C2(1), 0)), S.Half)) == (Rational(1, 4) - exp(-2) / 4) * (exp(-2) / 2 + S.Half) assert (P( Not(Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 2)) | (Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 2)), Eq(P(Eq(C2(1), 0)), Rational(1, 4)) & Eq(P(Eq(C2(1), 1)), Rational(1, 4)), ) is S.One) assert (E(C2(Rational(3, 2)), Eq(C2(0), 2)) == -exp(-3) / 2 + 2 * exp(Rational(-3, 2)) + S.Half) assert variance(C2(Rational(3, 2)), Eq( C2(0), 1)) == ((S.Half - exp(-3) / 2)**2 * (exp(-3) / 2 + S.Half) + (Rational(-1, 2) - exp(-3) / 2)**2 * (S.Half - exp(-3) / 2)) raises(KeyError, lambda: P(Eq(C2(1), 0), Eq(P(Eq(C2(1), 1)), S.Half))) assert P(Eq(C2(1), 0), Eq(P(Eq(C2(5), 1)), S.Half)) == Probability(Eq(C2(1), 0)) TS1 = MatrixSymbol("G", 3, 3) CS1 = ContinuousMarkovChain("C", [0, 1, 2], TS1) A = CS1.generator_matrix assert CS1.transition_probabilities(A)(t) == exp(t * A)
def test_issue_12237(): X = Normal('X', 0, 1) Y = Normal('Y', 0, 1) U = P(X > 0, X) V = P(Y < 0, X) assert U == Probability(X > 0, X) assert str(V) == '1/2'
def test_WienerProcess(): X = WienerProcess("X") assert X.state_space == S.Reals assert X.index_set == Interval(0, oo) t, d, x, y = symbols('t d x y', positive=True) assert isinstance(X(t), RandomIndexedSymbol) assert X.distribution(t) == NormalDistribution(0, sqrt(t)) with warns_deprecated_sympy(): X.distribution(X(t)) raises(ValueError, lambda: PoissonProcess("X", -1)) raises(NotImplementedError, lambda: X[t]) raises(IndexError, lambda: X(-2)) assert X.joint_distribution(X(2), X(3)) == JointDistributionHandmade( Lambda((X(2), X(3)), sqrt(6) * exp(-X(2)**2 / 4) * exp(-X(3)**2 / 6) / (12 * pi))) assert X.joint_distribution(4, 6) == JointDistributionHandmade( Lambda((X(4), X(6)), sqrt(6) * exp(-X(4)**2 / 8) * exp(-X(6)**2 / 12) / (24 * pi))) assert P(X(t) < 3).simplify() == erf(3 * sqrt(2) / (2 * sqrt(t))) / 2 + S(1) / 2 assert P(X(t) > 2, Contains(t, Interval.Lopen(3, 7))).simplify() == S(1)/2 -\ erf(sqrt(2)/2)/2 # Equivalent to P(X(1)>1)**4 assert P((X(t) > 4) & (X(d) > 3) & (X(x) > 2) & (X(y) > 1), Contains(t, Interval.Lopen(0, 1)) & Contains(d, Interval.Lopen(1, 2)) & Contains(x, Interval.Lopen(2, 3)) & Contains(y, Interval.Lopen(3, 4))).simplify() ==\ (1 - erf(sqrt(2)/2))*(1 - erf(sqrt(2)))*(1 - erf(3*sqrt(2)/2))*(1 - erf(2*sqrt(2)))/16 # Contains an overlapping interval so, return Probability assert P((X(t) < 2) & (X(d) > 3), Contains(t, Interval.Lopen(0, 2)) & Contains(d, Interval.Ropen(2, 4))) == Probability( (X(d) > 3) & (X(t) < 2), Contains(d, Interval.Ropen(2, 4)) & Contains(t, Interval.Lopen(0, 2))) assert str(P(Not((X(t) < 5) & (X(d) > 3)), Contains(t, Interval.Ropen(2, 4)) & Contains(d, Interval.Lopen(7, 8))).simplify()) == \ '-(1 - erf(3*sqrt(2)/2))*(2 - erfc(5/2))/4 + 1' # Distribution has mean 0 at each timestamp assert E(X(t)) == 0 assert E( x * (X(t) + X(d)) * (X(t)**2 + X(d)**2), Contains(t, Interval.Lopen(0, 1)) & Contains(d, Interval.Ropen(1, 2))) == Expectation( x * (X(d) + X(t)) * (X(d)**2 + X(t)**2), Contains(d, Interval.Ropen(1, 2)) & Contains(t, Interval.Lopen(0, 1))) assert E(X(t) + x * E(X(3))) == 0 #test issue 20078 assert (2 * X(t) + 3 * X(t)).simplify() == 5 * X(t) assert (2 * X(t) - 3 * X(t)).simplify() == -X(t) assert (2 * (0.25 * X(t))).simplify() == 0.5 * X(t) assert (2 * X(t) * 0.25 * X(t)).simplify() == 0.5 * X(t)**2 assert (X(t)**2 + X(t)**3).simplify() == (X(t) + 1) * X(t)**2
def probability(condition, given_condition=None, numsamples=None, evaluate=True, **kwargs): """ Probability that a condition is true, optionally given a second condition Parameters ========== condition : Combination of Relationals containing RandomSymbols The condition of which you want to compute the probability given_condition : Combination of Relationals containing RandomSymbols A conditional expression. P(X > 1, X > 0) is expectation of X > 1 given X > 0 numsamples : int Enables sampling and approximates the probability with this many samples evaluate : Bool (defaults to True) In case of continuous systems return unevaluated integral Examples ======== >>> from sympy.stats import P, Die >>> from sympy import Eq >>> X, Y = Die('X', 6), Die('Y', 6) >>> P(X > 3) 1/2 >>> P(Eq(X, 5), X > 2) # Probability that X == 5 given that X > 2 1/4 >>> P(X > Y) 5/12 """ kwargs['numsamples'] = numsamples from sympy.stats.symbolic_probability import Probability if evaluate: return Probability(condition, given_condition).doit(**kwargs) ### TODO: Remove the user warnings in the future releases message = ( "Since version 1.7, using `evaluate=False` returns `Probability` " "object. If you want unevaluated Integral/Sum use " "`P(condition, given_condition, evaluate=False).rewrite(Integral)`") warnings.warn(filldedent(message)) return Probability(condition, given_condition)
def test_issue_12237(): X = Normal('X', 0, 1) Y = Normal('Y', 0, 1) U = P(X > 0, X) V = P(Y < 0, X) W = P(X + Y > 0, X) assert W == Probability(X + Y > 0, X) assert U == BernoulliDistribution(S(1) / 2, S(0), S(1)) assert str(V) == '1/2'
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 probability(self, condition): _domain = self.restricted_domain(condition) if condition == False or _domain is S.EmptySet: return S.Zero if condition == True or _domain == self.set: return S.One try: return self.eval_prob(_domain) except NotImplementedError: return Probability(condition)
def probability(condition, given_condition=None, numsamples=None, evaluate=True, **kwargs): """ Probability that a condition is true, optionally given a second condition Parameters ========== condition : Combination of Relationals containing RandomSymbols The condition of which you want to compute the probability given_condition : Combination of Relationals containing RandomSymbols A conditional expression. P(X > 1, X > 0) is expectation of X > 1 given X > 0 numsamples : int Enables sampling and approximates the probability with this many samples evaluate : Bool (defaults to True) In case of continuous systems return unevaluated integral Examples ======== >>> from sympy.stats import P, Die >>> from sympy import Eq >>> X, Y = Die('X', 6), Die('Y', 6) >>> P(X > 3) 1/2 >>> P(Eq(X, 5), X > 2) # Probability that X == 5 given that X > 2 1/4 >>> P(X > Y) 5/12 """ kwargs['numsamples'] = numsamples from sympy.stats.symbolic_probability import Probability if evaluate: return Probability(condition, given_condition).doit(**kwargs) return Probability(condition, given_condition).rewrite( Integral) # will return Sum in case of discrete RV
def probability(self, condition): complement = isinstance(condition, Ne) if complement: condition = Eq(condition.args[0], condition.args[1]) _domain = self.where(condition).set if condition == False or _domain is S.EmptySet: return S.Zero if condition == True or _domain == self.set: return S.One prob = self.eval_prob(_domain) if prob == None: prob = Probability(condition) return prob if not complement else S.One - prob
def probability(self, condition): complement = isinstance(condition, Ne) if complement: condition = Eq(condition.args[0], condition.args[1]) _domain = self.restricted_domain(condition) if condition == False or _domain is S.EmptySet: return S.Zero if condition == True or _domain == self.set: return S.One try: prob = self.eval_prob(_domain) return prob if not complement else S.One - prob except NotImplementedError: return Probability(condition)
def probability(self, condition): complement = isinstance(condition, Ne) if complement: condition = Eq(condition.args[0], condition.args[1]) try: _domain = self.where(condition).set if condition == False or _domain is S.EmptySet: return S.Zero if condition == True or _domain == self.domain.set: return S.One prob = self.eval_prob(_domain) except NotImplementedError: from sympy.stats.rv import density expr = condition.lhs - condition.rhs dens = density(expr) if not isinstance(dens, DiscreteDistribution): dens = DiscreteDistributionHandmade(dens) z = Dummy('z', real = True) space = SingleDiscretePSpace(z, dens) prob = space.probability(condition.__class__(space.value, 0)) if (prob == None): prob = Probability(condition) return prob if not complement else S.One - prob
def test_DiscreteMarkovChain(): # pass only the name X = DiscreteMarkovChain("X") assert isinstance(X.state_space, Range) assert X.index_set == S.Naturals0 assert isinstance(X.transition_probabilities, MatrixSymbol) t = symbols('t', positive=True, integer=True) assert isinstance(X[t], RandomIndexedSymbol) assert E(X[0]) == Expectation(X[0]) raises(TypeError, lambda: DiscreteMarkovChain(1)) raises(NotImplementedError, lambda: X(t)) raises(NotImplementedError, lambda: X.communication_classes()) raises(NotImplementedError, lambda: X.canonical_form()) raises(NotImplementedError, lambda: X.decompose()) nz = Symbol('n', integer=True) TZ = MatrixSymbol('M', nz, nz) SZ = Range(nz) YZ = DiscreteMarkovChain('Y', SZ, TZ) assert P(Eq(YZ[2], 1), Eq(YZ[1], 0)) == TZ[0, 1] raises(ValueError, lambda: sample_stochastic_process(t)) raises(ValueError, lambda: next(sample_stochastic_process(X))) # pass name and state_space # any hashable object should be a valid state # states should be valid as a tuple/set/list/Tuple/Range sym, rainy, cloudy, sunny = symbols('a Rainy Cloudy Sunny', real=True) state_spaces = [(1, 2, 3), [Str('Hello'), sym, DiscreteMarkovChain], Tuple(S(1), exp(sym), Str('World'), sympify=False), Range(-1, 5, 2), [rainy, cloudy, sunny]] chains = [ DiscreteMarkovChain("Y", state_space) for state_space in state_spaces ] for i, Y in enumerate(chains): assert isinstance(Y.transition_probabilities, MatrixSymbol) assert Y.state_space == state_spaces[i] or Y.state_space == FiniteSet( *state_spaces[i]) assert Y.number_of_states == 3 with ignore_warnings( UserWarning): # TODO: Restore tests once warnings are removed assert P(Eq(Y[2], 1), Eq(Y[0], 2), evaluate=False) == Probability(Eq(Y[2], 1), Eq(Y[0], 2)) assert E(Y[0]) == Expectation(Y[0]) raises(ValueError, lambda: next(sample_stochastic_process(Y))) raises(TypeError, lambda: DiscreteMarkovChain("Y", dict((1, 1)))) Y = DiscreteMarkovChain("Y", Range(1, t, 2)) assert Y.number_of_states == ceiling((t - 1) / 2) # pass name and transition_probabilities chains = [ DiscreteMarkovChain("Y", trans_probs=Matrix([[]])), DiscreteMarkovChain("Y", trans_probs=Matrix([[0, 1], [1, 0]])), DiscreteMarkovChain("Y", trans_probs=Matrix([[pi, 1 - pi], [sym, 1 - sym]])) ] for Z in chains: assert Z.number_of_states == Z.transition_probabilities.shape[0] assert isinstance(Z.transition_probabilities, ImmutableMatrix) # pass name, state_space and transition_probabilities T = Matrix([[0.5, 0.2, 0.3], [0.2, 0.5, 0.3], [0.2, 0.3, 0.5]]) TS = MatrixSymbol('T', 3, 3) Y = DiscreteMarkovChain("Y", [0, 1, 2], T) YS = DiscreteMarkovChain("Y", ['One', 'Two', 3], TS) assert Y.joint_distribution(1, Y[2], 3) == JointDistribution(Y[1], Y[2], Y[3]) raises(ValueError, lambda: Y.joint_distribution(Y[1].symbol, Y[2].symbol)) assert P(Eq(Y[3], 2), Eq(Y[1], 1)).round(2) == Float(0.36, 2) assert (P(Eq(YS[3], 2), Eq(YS[1], 1)) - (TS[0, 2] * TS[1, 0] + TS[1, 1] * TS[1, 2] + TS[1, 2] * TS[2, 2])).simplify() == 0 assert P(Eq(YS[1], 1), Eq(YS[2], 2)) == Probability(Eq(YS[1], 1)) assert P(Eq(YS[3], 3), Eq( YS[1], 1)) == TS[0, 2] * TS[1, 0] + TS[1, 1] * TS[1, 2] + TS[1, 2] * TS[2, 2] TO = Matrix([[0.25, 0.75, 0], [0, 0.25, 0.75], [0.75, 0, 0.25]]) assert P(Eq(Y[3], 2), Eq(Y[1], 1) & TransitionMatrixOf(Y, TO)).round(3) == Float( 0.375, 3) with ignore_warnings( UserWarning): ### TODO: Restore tests once warnings are removed assert E(Y[3], evaluate=False) == Expectation(Y[3]) assert E(Y[3], Eq(Y[2], 1)).round(2) == Float(1.1, 3) TSO = MatrixSymbol('T', 4, 4) raises( ValueError, lambda: str(P(Eq(YS[3], 2), Eq(YS[1], 1) & TransitionMatrixOf(YS, TSO)))) raises(TypeError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], symbols('M'))) raises( ValueError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], MatrixSymbol('T', 3, 4))) raises(ValueError, lambda: E(Y[3], Eq(Y[2], 6))) raises(ValueError, lambda: E(Y[2], Eq(Y[3], 1))) # extended tests for probability queries TO1 = Matrix([[Rational(1, 4), Rational(3, 4), 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) assert P( And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Probability(Eq(Y[0], 0)), Rational(1, 4)) & TransitionMatrixOf(Y, TO1)) == Rational(1, 16) assert P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), TransitionMatrixOf(Y, TO1)) == \ Probability(Eq(Y[0], 0))/4 assert P( Lt(X[1], 2) & Gt(X[1], 0), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1)) == Rational(1, 4) assert P( Lt(X[1], 2) & Gt(X[1], 0), Eq(X[0], 2) & StochasticStateSpaceOf(X, [None, 'None', 1]) & TransitionMatrixOf(X, TO1)) == Rational(1, 4) assert P( Ne(X[1], 2) & Ne(X[1], 1), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1)) is S.Zero assert P( Ne(X[1], 2) & Ne(X[1], 1), Eq(X[0], 2) & StochasticStateSpaceOf(X, [None, 'None', 1]) & TransitionMatrixOf(X, TO1)) is S.Zero assert P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Y[1], 1)) == 0.1 * Probability(Eq(Y[0], 0)) # testing properties of Markov chain TO2 = Matrix([[S.One, 0, 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) TO3 = Matrix([[Rational(1, 4), Rational(3, 4), 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) Y2 = DiscreteMarkovChain('Y', trans_probs=TO2) Y3 = DiscreteMarkovChain('Y', trans_probs=TO3) assert Y3.fundamental_matrix() == ImmutableMatrix( [[176, 81, -132], [36, 141, -52], [-44, -39, 208]]) / 125 assert Y2.is_absorbing_chain() == True assert Y3.is_absorbing_chain() == False assert Y2.canonical_form() == ([0, 1, 2], TO2) assert Y3.canonical_form() == ([0, 1, 2], TO3) assert Y2.decompose() == ([0, 1, 2], TO2[0:1, 0:1], TO2[1:3, 0:1], TO2[1:3, 1:3]) assert Y3.decompose() == ([0, 1, 2], TO3, Matrix(0, 3, []), Matrix(0, 0, [])) TO4 = Matrix([[Rational(1, 5), Rational(2, 5), Rational(2, 5)], [Rational(1, 10), S.Half, Rational(2, 5)], [Rational(3, 5), Rational(3, 10), Rational(1, 10)]]) Y4 = DiscreteMarkovChain('Y', trans_probs=TO4) w = ImmutableMatrix([[Rational(11, 39), Rational(16, 39), Rational(4, 13)]]) assert Y4.limiting_distribution == w assert Y4.is_regular() == True assert Y4.is_ergodic() == True TS1 = MatrixSymbol('T', 3, 3) Y5 = DiscreteMarkovChain('Y', trans_probs=TS1) assert Y5.limiting_distribution(w, TO4).doit() == True assert Y5.stationary_distribution(condition_set=True).subs( TS1, TO4).contains(w).doit() == S.true TO6 = Matrix([[S.One, 0, 0, 0, 0], [S.Half, 0, S.Half, 0, 0], [0, S.Half, 0, S.Half, 0], [0, 0, S.Half, 0, S.Half], [0, 0, 0, 0, 1]]) Y6 = DiscreteMarkovChain('Y', trans_probs=TO6) assert Y6.fundamental_matrix() == ImmutableMatrix( [[Rational(3, 2), S.One, S.Half], [S.One, S(2), S.One], [S.Half, S.One, Rational(3, 2)]]) assert Y6.absorbing_probabilities() == ImmutableMatrix( [[Rational(3, 4), Rational(1, 4)], [S.Half, S.Half], [Rational(1, 4), Rational(3, 4)]]) TO7 = Matrix([[Rational(1, 2), Rational(1, 4), Rational(1, 4)], [Rational(1, 2), 0, Rational(1, 2)], [Rational(1, 4), Rational(1, 4), Rational(1, 2)]]) Y7 = DiscreteMarkovChain('Y', trans_probs=TO7) assert Y7.is_absorbing_chain() == False assert Y7.fundamental_matrix() == ImmutableMatrix( [[Rational(86, 75), Rational(1, 25), Rational(-14, 75)], [Rational(2, 25), Rational(21, 25), Rational(2, 25)], [Rational(-14, 75), Rational(1, 25), Rational(86, 75)]]) # test for zero-sized matrix functionality X = DiscreteMarkovChain('X', trans_probs=Matrix([[]])) assert X.number_of_states == 0 assert X.stationary_distribution() == Matrix([[]]) assert X.communication_classes() == [] assert X.canonical_form() == ([], Matrix([[]])) assert X.decompose() == ([], Matrix([[]]), Matrix([[]]), Matrix([[]])) assert X.is_regular() == False assert X.is_ergodic() == False # test communication_class # see https://drive.google.com/drive/folders/1HbxLlwwn2b3U8Lj7eb_ASIUb5vYaNIjg?usp=sharing # tutorial 2.pdf TO7 = Matrix([[0, 5, 5, 0, 0], [0, 0, 0, 10, 0], [5, 0, 5, 0, 0], [0, 10, 0, 0, 0], [0, 3, 0, 3, 4]]) / 10 Y7 = DiscreteMarkovChain('Y', trans_probs=TO7) tuples = Y7.communication_classes() classes, recurrence, periods = list(zip(*tuples)) assert classes == ([1, 3], [0, 2], [4]) assert recurrence == (True, False, False) assert periods == (2, 1, 1) TO8 = Matrix([[0, 0, 0, 10, 0, 0], [5, 0, 5, 0, 0, 0], [0, 4, 0, 0, 0, 6], [10, 0, 0, 0, 0, 0], [0, 10, 0, 0, 0, 0], [0, 0, 0, 5, 5, 0] ]) / 10 Y8 = DiscreteMarkovChain('Y', trans_probs=TO8) tuples = Y8.communication_classes() classes, recurrence, periods = list(zip(*tuples)) assert classes == ([0, 3], [1, 2, 5, 4]) assert recurrence == (True, False) assert periods == (2, 2) TO9 = Matrix( [[2, 0, 0, 3, 0, 0, 3, 2, 0, 0], [0, 10, 0, 0, 0, 0, 0, 0, 0, 0], [0, 2, 2, 0, 0, 0, 0, 0, 3, 3], [0, 0, 0, 3, 0, 0, 6, 1, 0, 0], [0, 0, 0, 0, 5, 5, 0, 0, 0, 0], [0, 0, 0, 0, 0, 10, 0, 0, 0, 0], [4, 0, 0, 5, 0, 0, 1, 0, 0, 0], [2, 0, 0, 4, 0, 0, 2, 2, 0, 0], [3, 0, 1, 0, 0, 0, 0, 0, 4, 2], [0, 0, 4, 0, 0, 0, 0, 0, 3, 3]]) / 10 Y9 = DiscreteMarkovChain('Y', trans_probs=TO9) tuples = Y9.communication_classes() classes, recurrence, periods = list(zip(*tuples)) assert classes == ([0, 3, 6, 7], [1], [2, 8, 9], [5], [4]) assert recurrence == (True, True, False, True, False) assert periods == (1, 1, 1, 1, 1) # test canonical form # see https://www.dartmouth.edu/~chance/teaching_aids/books_articles/probability_book/Chapter11.pdf # example 11.13 T = Matrix([[1, 0, 0, 0, 0], [S(1) / 2, 0, S(1) / 2, 0, 0], [0, S(1) / 2, 0, S(1) / 2, 0], [0, 0, S(1) / 2, 0, S(1) / 2], [0, 0, 0, 0, S(1)]]) DW = DiscreteMarkovChain('DW', [0, 1, 2, 3, 4], T) states, A, B, C = DW.decompose() assert states == [0, 4, 1, 2, 3] assert A == Matrix([[1, 0], [0, 1]]) assert B == Matrix([[S(1) / 2, 0], [0, 0], [0, S(1) / 2]]) assert C == Matrix([[0, S(1) / 2, 0], [S(1) / 2, 0, S(1) / 2], [0, S(1) / 2, 0]]) states, new_matrix = DW.canonical_form() assert states == [0, 4, 1, 2, 3] assert new_matrix == Matrix([[1, 0, 0, 0, 0], [0, 1, 0, 0, 0], [S(1) / 2, 0, 0, S(1) / 2, 0], [0, 0, S(1) / 2, 0, S(1) / 2], [0, S(1) / 2, 0, S(1) / 2, 0]]) # test regular and ergodic # https://www.dartmouth.edu/~chance/teaching_aids/books_articles/probability_book/Chapter11.pdf T = Matrix([[0, 4, 0, 0, 0], [1, 0, 3, 0, 0], [0, 2, 0, 2, 0], [0, 0, 3, 0, 1], [0, 0, 0, 4, 0]]) / 4 X = DiscreteMarkovChain('X', trans_probs=T) assert not X.is_regular() assert X.is_ergodic() T = Matrix([[0, 1], [1, 0]]) X = DiscreteMarkovChain('X', trans_probs=T) assert not X.is_regular() assert X.is_ergodic() # http://www.math.wisc.edu/~valko/courses/331/MC2.pdf T = Matrix([[2, 1, 1], [2, 0, 2], [1, 1, 2]]) / 4 X = DiscreteMarkovChain('X', trans_probs=T) assert X.is_regular() assert X.is_ergodic() # https://docs.ufpr.br/~lucambio/CE222/1S2014/Kemeny-Snell1976.pdf T = Matrix([[1, 1], [1, 1]]) / 2 X = DiscreteMarkovChain('X', trans_probs=T) assert X.is_regular() assert X.is_ergodic() # test is_absorbing_chain T = Matrix([[0, 1, 0], [1, 0, 0], [0, 0, 1]]) X = DiscreteMarkovChain('X', trans_probs=T) assert not X.is_absorbing_chain() # https://en.wikipedia.org/wiki/Absorbing_Markov_chain T = Matrix([[1, 1, 0, 0], [0, 1, 1, 0], [1, 0, 0, 1], [0, 0, 0, 2]]) / 2 X = DiscreteMarkovChain('X', trans_probs=T) assert X.is_absorbing_chain() T = Matrix([[2, 0, 0, 0, 0], [1, 0, 1, 0, 0], [0, 1, 0, 1, 0], [0, 0, 1, 0, 1], [0, 0, 0, 0, 2]]) / 2 X = DiscreteMarkovChain('X', trans_probs=T) assert X.is_absorbing_chain() # test custom state space Y10 = DiscreteMarkovChain('Y', [1, 2, 3], TO2) tuples = Y10.communication_classes() classes, recurrence, periods = list(zip(*tuples)) assert classes == ([1], [2, 3]) assert recurrence == (True, False) assert periods == (1, 1) assert Y10.canonical_form() == ([1, 2, 3], TO2) assert Y10.decompose() == ([1, 2, 3], TO2[0:1, 0:1], TO2[1:3, 0:1], TO2[1:3, 1:3]) # testing miscellaneous queries T = Matrix([[S.Half, Rational(1, 4), Rational(1, 4)], [Rational(1, 3), 0, Rational(2, 3)], [S.Half, S.Half, 0]]) X = DiscreteMarkovChain('X', [0, 1, 2], T) assert P( Eq(X[1], 2) & Eq(X[2], 1) & Eq(X[3], 0), Eq(P(Eq(X[1], 0)), Rational(1, 4)) & Eq(P(Eq(X[1], 1)), Rational(1, 4))) == Rational(1, 12) assert P(Eq(X[2], 1) | Eq(X[2], 2), Eq(X[1], 1)) == Rational(2, 3) assert P(Eq(X[2], 1) & Eq(X[2], 2), Eq(X[1], 1)) is S.Zero assert P(Ne(X[2], 2), Eq(X[1], 1)) == Rational(1, 3) assert E(X[1]**2, Eq(X[0], 1)) == Rational(8, 3) assert variance(X[1], Eq(X[0], 1)) == Rational(8, 9) raises(ValueError, lambda: E(X[1], Eq(X[2], 1))) raises(ValueError, lambda: DiscreteMarkovChain('X', [0, 1], T)) # testing miscellaneous queries with different state space X = DiscreteMarkovChain('X', ['A', 'B', 'C'], T) assert P( Eq(X[1], 2) & Eq(X[2], 1) & Eq(X[3], 0), Eq(P(Eq(X[1], 0)), Rational(1, 4)) & Eq(P(Eq(X[1], 1)), Rational(1, 4))) == Rational(1, 12) assert P(Eq(X[2], 1) | Eq(X[2], 2), Eq(X[1], 1)) == Rational(2, 3) assert P(Eq(X[2], 1) & Eq(X[2], 2), Eq(X[1], 1)) is S.Zero assert P(Ne(X[2], 2), Eq(X[1], 1)) == Rational(1, 3) a = X.state_space.args[0] c = X.state_space.args[2] assert (E(X[1]**2, Eq(X[0], 1)) - (a**2 / 3 + 2 * c**2 / 3)).simplify() == 0 assert (variance(X[1], Eq(X[0], 1)) - (2 * (-a / 3 + c / 3)**2 / 3 + (2 * a / 3 - 2 * c / 3)**2 / 3)).simplify() == 0 raises(ValueError, lambda: E(X[1], Eq(X[2], 1))) #testing queries with multiple RandomIndexedSymbols T = Matrix([[Rational(5, 10), Rational(3, 10), Rational(2, 10)], [Rational(2, 10), Rational(7, 10), Rational(1, 10)], [Rational(3, 10), Rational(3, 10), Rational(4, 10)]]) Y = DiscreteMarkovChain("Y", [0, 1, 2], T) assert P(Eq(Y[7], Y[5]), Eq(Y[2], 0)).round(5) == Float(0.44428, 5) assert P(Gt(Y[3], Y[1]), Eq(Y[0], 0)).round(2) == Float(0.36, 2) assert P(Le(Y[5], Y[10]), Eq(Y[4], 2)).round(6) == Float(0.583120, 6) assert Float(P(Eq(Y[10], Y[5]), Eq(Y[4], 1)), 14) == Float(1 - P(Ne(Y[10], Y[5]), Eq(Y[4], 1)), 14) assert Float(P(Gt(Y[8], Y[9]), Eq(Y[3], 2)), 14) == Float(1 - P(Le(Y[8], Y[9]), Eq(Y[3], 2)), 14) assert Float(P(Lt(Y[1], Y[4]), Eq(Y[0], 0)), 14) == Float(1 - P(Ge(Y[1], Y[4]), Eq(Y[0], 0)), 14) assert P(Eq(Y[5], Y[10]), Eq(Y[2], 1)) == P(Eq(Y[10], Y[5]), Eq(Y[2], 1)) assert P(Gt(Y[1], Y[2]), Eq(Y[0], 1)) == P(Lt(Y[2], Y[1]), Eq(Y[0], 1)) assert P(Ge(Y[7], Y[6]), Eq(Y[4], 1)) == P(Le(Y[6], Y[7]), Eq(Y[4], 1)) #test symbolic queries a, b, c, d = symbols('a b c d') T = Matrix([[Rational(1, 10), Rational(4, 10), Rational(5, 10)], [Rational(3, 10), Rational(4, 10), Rational(3, 10)], [Rational(7, 10), Rational(2, 10), Rational(1, 10)]]) Y = DiscreteMarkovChain("Y", [0, 1, 2], T) query = P(Eq(Y[a], b), Eq(Y[c], d)) assert query.subs({ a: 10, b: 2, c: 5, d: 1 }).evalf().round(4) == P(Eq(Y[10], 2), Eq(Y[5], 1)).round(4) assert query.subs({ a: 15, b: 0, c: 10, d: 1 }).evalf().round(4) == P(Eq(Y[15], 0), Eq(Y[10], 1)).round(4) query_gt = P(Gt(Y[a], b), Eq(Y[c], d)) query_le = P(Le(Y[a], b), Eq(Y[c], d)) assert query_gt.subs({ a: 5, b: 2, c: 1, d: 0 }).evalf() + query_le.subs({ a: 5, b: 2, c: 1, d: 0 }).evalf() == 1 query_ge = P(Ge(Y[a], b), Eq(Y[c], d)) query_lt = P(Lt(Y[a], b), Eq(Y[c], d)) assert query_ge.subs({ a: 4, b: 1, c: 0, d: 2 }).evalf() + query_lt.subs({ a: 4, b: 1, c: 0, d: 2 }).evalf() == 1 #test issue 20078 assert (2 * Y[1] + 3 * Y[1]).simplify() == 5 * Y[1] assert (2 * Y[1] - 3 * Y[1]).simplify() == -Y[1] assert (2 * (0.25 * Y[1])).simplify() == 0.5 * Y[1] assert ((2 * Y[1]) * (0.25 * Y[1])).simplify() == 0.5 * Y[1]**2 assert (Y[1]**2 + Y[1]**3).simplify() == (Y[1] + 1) * Y[1]**2
def probability(condition, given_condition=None, numsamples=None, evaluate=True, **kwargs): """ Probability that a condition is true, optionally given a second condition Parameters ========== condition : Combination of Relationals containing RandomSymbols The condition of which you want to compute the probability given_condition : Combination of Relationals containing RandomSymbols A conditional expression. P(X > 1, X > 0) is expectation of X > 1 given X > 0 numsamples : int Enables sampling and approximates the probability with this many samples evaluate : Bool (defaults to True) In case of continuous systems return unevaluated integral Examples ======== >>> from sympy.stats import P, Die >>> from sympy import Eq >>> X, Y = Die('X', 6), Die('Y', 6) >>> P(X > 3) 1/2 >>> P(Eq(X, 5), X > 2) # Probability that X == 5 given that X > 2 1/4 >>> P(X > Y) 5/12 """ condition = sympify(condition) given_condition = sympify(given_condition) if condition.has(RandomIndexedSymbol): return pspace(condition).probability(condition, given_condition, evaluate, **kwargs) 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(probability(condition), 0, 1) if any([dependent(rv, given_condition) for rv in condrv]): from sympy.stats.symbolic_probability import Probability return Probability(condition, given_condition) else: return probability(condition) 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: 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 condition is S.false: return S.Zero if numsamples: return sampling_P(condition, given_condition, numsamples=numsamples, **kwargs) if given_condition is not None: # If there is a condition # Recompute on new conditional expr return probability(given(condition, given_condition, **kwargs), **kwargs) # Otherwise pass work off to the ProbabilitySpace result = pspace(condition).probability(condition, **kwargs) if evaluate and hasattr(result, 'doit'): return result.doit() else: return result
def probability(self, condition, given_condition=None, evaluate=True, **kwargs): """ Handles probability queries for Markov process. Parameters ========== condition: Relational given_condition: Relational/And Returns ======= Probability If the information is not sufficient. Expr In all other cases. 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, new_given_condition = \ self._preprocess(given_condition, evaluate) if check: return Probability(condition, new_given_condition) if isinstance(self, ContinuousMarkovChain): trans_probs = self.transition_probabilities(mat) elif isinstance(self, DiscreteMarkovChain): trans_probs = mat if isinstance(condition, Relational): rv, states = (list(condition.atoms(RandomIndexedSymbol))[0], condition.as_set()) if isinstance(new_given_condition, And): gcs = new_given_condition.args else: gcs = (new_given_condition, ) grvs = new_given_condition.atoms(RandomIndexedSymbol) min_key_rv = None for grv in grvs: if grv.key <= rv.key: min_key_rv = grv if min_key_rv == None: return Probability(condition) prob, gstate = dict(), None for gc in gcs: if gc.has(min_key_rv): if gc.has(Probability): p, gp = (gc.rhs, gc.lhs) if isinstance(gc.lhs, Probability) \ else (gc.lhs, gc.rhs) gr = gp.args[0] gset = Intersection(gr.as_set(), state_space) gstate = list(gset)[0] prob[gset] = p else: _, gstate = (gc.lhs.key, gc.rhs) if isinstance(gc.lhs, RandomIndexedSymbol) \ else (gc.rhs.key, gc.lhs) if any( (k not in self.index_set) for k in (rv.key, min_key_rv.key)): raise IndexError( "The timestamps of the process are not in it's index set.") states = Intersection(states, state_space) for state in Union(states, FiniteSet(gstate)): if Ge(state, mat.shape[0]) == True: raise IndexError( "No information is available for (%s, %s) in " "transition probabilities of shape, (%s, %s). " "State space is zero indexed." % (gstate, state, mat.shape[0], mat.shape[1])) if prob: gstates = Union(*prob.keys()) if len(gstates) == 1: gstate = list(gstates)[0] gprob = list(prob.values())[0] prob[gstates] = gprob elif len(gstates) == len(state_space) - 1: gstate = list(state_space - gstates)[0] gprob = S(1) - sum(prob.values()) prob[state_space - gstates] = gprob else: raise ValueError("Conflicting information.") else: gprob = S(1) if min_key_rv == rv: return sum([prob[FiniteSet(state)] for state in states]) if isinstance(self, ContinuousMarkovChain): return gprob * sum([ trans_probs(rv.key - min_key_rv.key).__getitem__( (gstate, state)) for state in states ]) if isinstance(self, DiscreteMarkovChain): return gprob * sum([(trans_probs **(rv.key - min_key_rv.key)).__getitem__( (gstate, state)) for state in states]) if isinstance(condition, Not): expr = condition.args[0] return S(1) - self.probability(expr, given_condition, evaluate, ** kwargs) if isinstance(condition, And): compute_later, state2cond, conds = [], dict(), condition.args for expr in conds: if isinstance(expr, Relational): ris = list(expr.atoms(RandomIndexedSymbol))[0] if state2cond.get(ris, None) is None: state2cond[ris] = S.true state2cond[ris] &= expr else: compute_later.append(expr) ris = [] for ri in state2cond: ris.append(ri) cset = Intersection(state2cond[ri].as_set(), state_space) if len(cset) == 0: return S.Zero state2cond[ri] = cset.as_relational(ri) sorted_ris = sorted(ris, key=lambda ri: ri.key) prod = self.probability(state2cond[sorted_ris[0]], given_condition, evaluate, **kwargs) for i in range(1, len(sorted_ris)): ri, prev_ri = sorted_ris[i], sorted_ris[i - 1] if not isinstance(state2cond[ri], Eq): raise ValueError( "The process is in multiple states at %s, unable to determine the probability." % (ri)) mat_of = TransitionMatrixOf(self, mat) if isinstance( self, DiscreteMarkovChain) else GeneratorMatrixOf( self, mat) prod *= self.probability( state2cond[ri], state2cond[prev_ri] & mat_of & StochasticStateSpaceOf(self, state_space), evaluate, **kwargs) for expr in compute_later: prod *= self.probability(expr, given_condition, evaluate, **kwargs) return prod if isinstance(condition, Or): return sum([ self.probability(expr, given_condition, evaluate, **kwargs) for expr in condition.args ]) raise NotImplementedError( "Mechanism for handling (%s, %s) queries hasn't been " "implemented yet." % (expr, condition))
def test_DiscreteMarkovChain(): # pass only the name X = DiscreteMarkovChain("X") assert X.state_space == S.Reals assert X.index_set == S.Naturals0 assert X.transition_probabilities == None t = symbols("t", positive=True, integer=True) assert isinstance(X[t], RandomIndexedSymbol) assert E(X[0]) == Expectation(X[0]) raises(TypeError, lambda: DiscreteMarkovChain(1)) raises(NotImplementedError, lambda: X(t)) # pass name and state_space Y = DiscreteMarkovChain("Y", [1, 2, 3]) assert Y.transition_probabilities == None assert Y.state_space == FiniteSet(1, 2, 3) assert P(Eq(Y[2], 1), Eq(Y[0], 2)) == Probability(Eq(Y[2], 1), Eq(Y[0], 2)) assert E(X[0]) == Expectation(X[0]) raises(TypeError, lambda: DiscreteMarkovChain("Y", dict((1, 1)))) # pass name, state_space and transition_probabilities T = Matrix([[0.5, 0.2, 0.3], [0.2, 0.5, 0.3], [0.2, 0.3, 0.5]]) TS = MatrixSymbol("T", 3, 3) Y = DiscreteMarkovChain("Y", [0, 1, 2], T) YS = DiscreteMarkovChain("Y", [0, 1, 2], TS) assert YS._transient2transient() == None assert YS._transient2absorbing() == None assert Y.joint_distribution(1, Y[2], 3) == JointDistribution(Y[1], Y[2], Y[3]) raises(ValueError, lambda: Y.joint_distribution(Y[1].symbol, Y[2].symbol)) assert P(Eq(Y[3], 2), Eq(Y[1], 1)).round(2) == Float(0.36, 2) assert (str(P(Eq(YS[3], 2), Eq( YS[1], 1))) == "T[0, 2]*T[1, 0] + T[1, 1]*T[1, 2] + T[1, 2]*T[2, 2]") assert P(Eq(YS[1], 1), Eq(YS[2], 2)) == Probability(Eq(YS[1], 1)) assert P(Eq(YS[3], 3), Eq(YS[1], 1)) is S.Zero TO = Matrix([[0.25, 0.75, 0], [0, 0.25, 0.75], [0.75, 0, 0.25]]) assert P(Eq(Y[3], 2), Eq(Y[1], 1) & TransitionMatrixOf(Y, TO)).round(3) == Float( 0.375, 3) assert E(Y[3], evaluate=False) == Expectation(Y[3]) assert E(Y[3], Eq(Y[2], 1)).round(2) == Float(1.1, 3) TSO = MatrixSymbol("T", 4, 4) raises( ValueError, lambda: str(P(Eq(YS[3], 2), Eq(YS[1], 1) & TransitionMatrixOf(YS, TSO))), ) raises(TypeError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], symbols("M"))) raises( ValueError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], MatrixSymbol("T", 3, 4))) raises(ValueError, lambda: E(Y[3], Eq(Y[2], 6))) raises(ValueError, lambda: E(Y[2], Eq(Y[3], 1))) # extended tests for probability queries TO1 = Matrix([ [Rational(1, 4), Rational(3, 4), 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)], ]) assert P( And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Probability(Eq(Y[0], 0)), Rational(1, 4)) & TransitionMatrixOf(Y, TO1), ) == Rational(1, 16) assert (P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), TransitionMatrixOf(Y, TO1)) == Probability(Eq(Y[0], 0)) / 4) assert P( Lt(X[1], 2) & Gt(X[1], 0), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1), ) == Rational(1, 4) assert (P( Ne(X[1], 2) & Ne(X[1], 1), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1), ) is S.Zero) assert P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Y[1], 1)) == 0.1 * Probability(Eq(Y[0], 0)) # testing properties of Markov chain TO2 = Matrix([ [S.One, 0, 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)], ]) TO3 = Matrix([ [Rational(1, 4), Rational(3, 4), 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)], ]) Y2 = DiscreteMarkovChain("Y", trans_probs=TO2) Y3 = DiscreteMarkovChain("Y", trans_probs=TO3) assert Y3._transient2absorbing() == None raises(ValueError, lambda: Y3.fundamental_matrix()) assert Y2.is_absorbing_chain() == True assert Y3.is_absorbing_chain() == False TO4 = Matrix([ [Rational(1, 5), Rational(2, 5), Rational(2, 5)], [Rational(1, 10), S.Half, Rational(2, 5)], [Rational(3, 5), Rational(3, 10), Rational(1, 10)], ]) Y4 = DiscreteMarkovChain("Y", trans_probs=TO4) w = ImmutableMatrix([[Rational(11, 39), Rational(16, 39), Rational(4, 13)]]) assert Y4.limiting_distribution == w assert Y4.is_regular() == True TS1 = MatrixSymbol("T", 3, 3) Y5 = DiscreteMarkovChain("Y", trans_probs=TS1) assert Y5.limiting_distribution(w, TO4).doit() == True TO6 = Matrix([ [S.One, 0, 0, 0, 0], [S.Half, 0, S.Half, 0, 0], [0, S.Half, 0, S.Half, 0], [0, 0, S.Half, 0, S.Half], [0, 0, 0, 0, 1], ]) Y6 = DiscreteMarkovChain("Y", trans_probs=TO6) assert Y6._transient2absorbing() == ImmutableMatrix([[S.Half, 0], [0, 0], [0, S.Half]]) assert Y6._transient2transient() == ImmutableMatrix([[0, S.Half, 0], [S.Half, 0, S.Half], [0, S.Half, 0]]) assert Y6.fundamental_matrix() == ImmutableMatrix([ [Rational(3, 2), S.One, S.Half], [S.One, S(2), S.One], [S.Half, S.One, Rational(3, 2)], ]) assert Y6.absorbing_probabilites() == ImmutableMatrix([ [Rational(3, 4), Rational(1, 4)], [S.Half, S.Half], [Rational(1, 4), Rational(3, 4)], ]) # testing miscellaneous queries T = Matrix([ [S.Half, Rational(1, 4), Rational(1, 4)], [Rational(1, 3), 0, Rational(2, 3)], [S.Half, S.Half, 0], ]) X = DiscreteMarkovChain("X", [0, 1, 2], T) assert P( Eq(X[1], 2) & Eq(X[2], 1) & Eq(X[3], 0), Eq(P(Eq(X[1], 0)), Rational(1, 4)) & Eq(P(Eq(X[1], 1)), Rational(1, 4)), ) == Rational(1, 12) assert P(Eq(X[2], 1) | Eq(X[2], 2), Eq(X[1], 1)) == Rational(2, 3) assert P(Eq(X[2], 1) & Eq(X[2], 2), Eq(X[1], 1)) is S.Zero assert P(Ne(X[2], 2), Eq(X[1], 1)) == Rational(1, 3) assert E(X[1]**2, Eq(X[0], 1)) == Rational(8, 3) assert variance(X[1], Eq(X[0], 1)) == Rational(8, 9) raises(ValueError, lambda: E(X[1], Eq(X[2], 1)))
def test_DiscreteMarkovChain(): # pass only the name X = DiscreteMarkovChain("X") assert isinstance(X.state_space, Range) assert isinstance(X.index_of, Range) assert not X._is_numeric assert X.index_set == S.Naturals0 assert isinstance(X.transition_probabilities, MatrixSymbol) t = symbols('t', positive=True, integer=True) assert isinstance(X[t], RandomIndexedSymbol) assert E(X[0]) == Expectation(X[0]) raises(TypeError, lambda: DiscreteMarkovChain(1)) raises(NotImplementedError, lambda: X(t)) raises(ValueError, lambda: sample_stochastic_process(t)) raises(ValueError, lambda: next(sample_stochastic_process(X))) # pass name and state_space # any hashable object should be a valid state # states should be valid as a tuple/set/list/Tuple/Range sym = symbols('a', real=True) state_spaces = [(1, 2, 3), [Str('Hello'), sym, DiscreteMarkovChain], Tuple(1, exp(sym), Str('World'), sympify=False), Range(-1, 7, 2)] chains = [ DiscreteMarkovChain("Y", state_spaces[0]), DiscreteMarkovChain("Y", state_spaces[1]), DiscreteMarkovChain("Y", state_spaces[2]) ] for i, Y in enumerate(chains): assert isinstance(Y.transition_probabilities, MatrixSymbol) assert Y.state_space == Tuple(*state_spaces[i]) assert Y.number_of_states == 3 assert not Y._is_numeric # because no transition matrix is provided assert Y.index_of[state_spaces[i][0]] == 0 assert Y.index_of[state_spaces[i][1]] == 1 assert Y.index_of[state_spaces[i][2]] == 2 with ignore_warnings( UserWarning): # TODO: Restore tests once warnings are removed assert P(Eq(Y[2], 1), Eq(Y[0], 2), evaluate=False) == Probability(Eq(Y[2], 1), Eq(Y[0], 2)) assert E(Y[0]) == Expectation(Y[0]) raises(ValueError, lambda: next(sample_stochastic_process(Y))) raises(TypeError, lambda: DiscreteMarkovChain("Y", dict((1, 1)))) Y = DiscreteMarkovChain("Y", Range(1, t, 2)) assert Y.number_of_states == ceiling((t - 1) / 2) raises(NotImplementedError, lambda: Y.index_of) # pass name and transition_probabilities chains = [ DiscreteMarkovChain("Y", trans_probs=Matrix([[]])), DiscreteMarkovChain("Y", trans_probs=Matrix([[0, 1], [1, 0]])), DiscreteMarkovChain("Y", trans_probs=Matrix([[pi, 1 - pi], [sym, 1 - sym]])) ] for Z in chains: assert Z.number_of_states == Z.transition_probabilities.shape[0] assert isinstance(Z.transition_probabilities, ImmutableDenseMatrix) assert isinstance(Z.state_space, Tuple) assert Z._is_numeric # pass name, state_space and transition_probabilities T = Matrix([[0.5, 0.2, 0.3], [0.2, 0.5, 0.3], [0.2, 0.3, 0.5]]) TS = MatrixSymbol('T', 3, 3) Y = DiscreteMarkovChain("Y", [0, 1, 2], T) YS = DiscreteMarkovChain("Y", ['One', 'Two', 3], TS) assert YS._transient2transient() == None assert YS._transient2absorbing() == None assert Y.joint_distribution(1, Y[2], 3) == JointDistribution(Y[1], Y[2], Y[3]) raises(ValueError, lambda: Y.joint_distribution(Y[1].symbol, Y[2].symbol)) assert P(Eq(Y[3], 2), Eq(Y[1], 1)).round(2) == Float(0.36, 2) assert (P(Eq(YS[3], 2), Eq(YS[1], 1)) - (TS[0, 2] * TS[1, 0] + TS[1, 1] * TS[1, 2] + TS[1, 2] * TS[2, 2])).simplify() == 0 assert P(Eq(YS[1], 1), Eq(YS[2], 2)) == Probability(Eq(YS[1], 1)) assert P(Eq(YS[3], 3), Eq(YS[1], 1)) is S.Zero TO = Matrix([[0.25, 0.75, 0], [0, 0.25, 0.75], [0.75, 0, 0.25]]) assert P(Eq(Y[3], 2), Eq(Y[1], 1) & TransitionMatrixOf(Y, TO)).round(3) == Float( 0.375, 3) with ignore_warnings( UserWarning): ### TODO: Restore tests once warnings are removed assert E(Y[3], evaluate=False) == Expectation(Y[3]) assert E(Y[3], Eq(Y[2], 1)).round(2) == Float(1.1, 3) TSO = MatrixSymbol('T', 4, 4) raises( ValueError, lambda: str(P(Eq(YS[3], 2), Eq(YS[1], 1) & TransitionMatrixOf(YS, TSO)))) raises(TypeError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], symbols('M'))) raises( ValueError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], MatrixSymbol('T', 3, 4))) raises(ValueError, lambda: E(Y[3], Eq(Y[2], 6))) raises(ValueError, lambda: E(Y[2], Eq(Y[3], 1))) # extended tests for probability queries TO1 = Matrix([[Rational(1, 4), Rational(3, 4), 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) assert P( And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Probability(Eq(Y[0], 0)), Rational(1, 4)) & TransitionMatrixOf(Y, TO1)) == Rational(1, 16) assert P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), TransitionMatrixOf(Y, TO1)) == \ Probability(Eq(Y[0], 0))/4 assert P( Lt(X[1], 2) & Gt(X[1], 0), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1)) == Rational(1, 4) assert P( Lt(X[1], 2) & Gt(X[1], 0), Eq(X[0], 2) & StochasticStateSpaceOf(X, [None, 'None', 1]) & TransitionMatrixOf(X, TO1)) == Rational(1, 4) assert P( Ne(X[1], 2) & Ne(X[1], 1), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1)) is S.Zero assert P( Ne(X[1], 2) & Ne(X[1], 1), Eq(X[0], 2) & StochasticStateSpaceOf(X, [None, 'None', 1]) & TransitionMatrixOf(X, TO1)) is S.Zero assert P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Y[1], 1)) == 0.1 * Probability(Eq(Y[0], 0)) # testing properties of Markov chain TO2 = Matrix([[S.One, 0, 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) TO3 = Matrix([[Rational(1, 4), Rational(3, 4), 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) Y2 = DiscreteMarkovChain('Y', trans_probs=TO2) Y3 = DiscreteMarkovChain('Y', trans_probs=TO3) assert Y3._transient2absorbing() == None raises(ValueError, lambda: Y3.fundamental_matrix()) assert Y2.is_absorbing_chain() == True assert Y3.is_absorbing_chain() == False TO4 = Matrix([[Rational(1, 5), Rational(2, 5), Rational(2, 5)], [Rational(1, 10), S.Half, Rational(2, 5)], [Rational(3, 5), Rational(3, 10), Rational(1, 10)]]) Y4 = DiscreteMarkovChain('Y', trans_probs=TO4) w = ImmutableMatrix([[Rational(11, 39), Rational(16, 39), Rational(4, 13)]]) assert Y4.limiting_distribution == w assert Y4.is_regular() == True TS1 = MatrixSymbol('T', 3, 3) Y5 = DiscreteMarkovChain('Y', trans_probs=TS1) assert Y5.limiting_distribution(w, TO4).doit() == True TO6 = Matrix([[S.One, 0, 0, 0, 0], [S.Half, 0, S.Half, 0, 0], [0, S.Half, 0, S.Half, 0], [0, 0, S.Half, 0, S.Half], [0, 0, 0, 0, 1]]) Y6 = DiscreteMarkovChain('Y', trans_probs=TO6) assert Y6._transient2absorbing() == ImmutableMatrix([[S.Half, 0], [0, 0], [0, S.Half]]) assert Y6._transient2transient() == ImmutableMatrix([[0, S.Half, 0], [S.Half, 0, S.Half], [0, S.Half, 0]]) assert Y6.fundamental_matrix() == ImmutableMatrix( [[Rational(3, 2), S.One, S.Half], [S.One, S(2), S.One], [S.Half, S.One, Rational(3, 2)]]) assert Y6.absorbing_probabilities() == ImmutableMatrix( [[Rational(3, 4), Rational(1, 4)], [S.Half, S.Half], [Rational(1, 4), Rational(3, 4)]]) # testing miscellaneous queries T = Matrix([[S.Half, Rational(1, 4), Rational(1, 4)], [Rational(1, 3), 0, Rational(2, 3)], [S.Half, S.Half, 0]]) X = DiscreteMarkovChain('X', [0, 1, 2], T) assert P( Eq(X[1], 2) & Eq(X[2], 1) & Eq(X[3], 0), Eq(P(Eq(X[1], 0)), Rational(1, 4)) & Eq(P(Eq(X[1], 1)), Rational(1, 4))) == Rational(1, 12) assert P(Eq(X[2], 1) | Eq(X[2], 2), Eq(X[1], 1)) == Rational(2, 3) assert P(Eq(X[2], 1) & Eq(X[2], 2), Eq(X[1], 1)) is S.Zero assert P(Ne(X[2], 2), Eq(X[1], 1)) == Rational(1, 3) assert E(X[1]**2, Eq(X[0], 1)) == Rational(8, 3) assert variance(X[1], Eq(X[0], 1)) == Rational(8, 9) raises(ValueError, lambda: E(X[1], Eq(X[2], 1))) raises(ValueError, lambda: DiscreteMarkovChain('X', [0, 1], T)) # testing miscellaneous queries with different state space X = DiscreteMarkovChain('X', ['A', 'B', 'C'], T) assert P( Eq(X[1], 2) & Eq(X[2], 1) & Eq(X[3], 0), Eq(P(Eq(X[1], 0)), Rational(1, 4)) & Eq(P(Eq(X[1], 1)), Rational(1, 4))) == Rational(1, 12) assert P(Eq(X[2], 1) | Eq(X[2], 2), Eq(X[1], 1)) == Rational(2, 3) assert P(Eq(X[2], 1) & Eq(X[2], 2), Eq(X[1], 1)) is S.Zero assert P(Ne(X[2], 2), Eq(X[1], 1)) == Rational(1, 3) a = X.state_space.args[0] c = X.state_space.args[2] assert (E(X[1]**2, Eq(X[0], 1)) - (a**2 / 3 + 2 * c**2 / 3)).simplify() == 0 assert (variance(X[1], Eq(X[0], 1)) - (2 * (-a / 3 + c / 3)**2 / 3 + (2 * a / 3 - 2 * c / 3)**2 / 3)).simplify() == 0 raises(ValueError, lambda: E(X[1], Eq(X[2], 1)))
def probability(self, condition, given_condition=None, evaluate=True, **kwargs): """ Handles probability queries for discrete Markov chains. Parameters ========== condition: Relational given_condition: Relational/And Returns ======= Probability If the transition probabilities are not available Expr If the transition probabilities is MatrixSymbol or Matrix 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 and state space using StochasticStateSpaceOf in given_condition using & or And. """ check, trans_probs, state_space, given_condition = \ self._preprocess(given_condition, evaluate) if check: return Probability(condition, given_condition) if isinstance(condition, Eq) and \ isinstance(given_condition, Eq) and \ len(given_condition.atoms(RandomSymbol)) == 1: # handles simple queries like P(Eq(X[i], dest_state), Eq(X[i], init_state)) lhsc, rhsc = condition.lhs, condition.rhs lhsg, rhsg = given_condition.lhs, given_condition.rhs if not isinstance(lhsc, RandomIndexedSymbol): lhsc, rhsc = (rhsc, lhsc) if not isinstance(lhsg, RandomIndexedSymbol): lhsg, rhsg = (rhsg, lhsg) keyc, statec, keyg, stateg = (lhsc.key, rhsc, lhsg.key, rhsg) if Lt(stateg, trans_probs.shape[0]) == False or Lt( statec, trans_probs.shape[1]) == False: raise IndexError( "No information is available for (%s, %s) in " "transition probabilities of shape, (%s, %s). " "State space is zero indexed." % (stateg, statec, trans_probs.shape[0], trans_probs.shape[1])) if keyc < keyg: raise ValueError( "Incorrect given condition is given, probability " "of past state cannot be computed from future state.") nsteptp = trans_probs**(keyc - keyg) if hasattr(nsteptp, "__getitem__"): return nsteptp.__getitem__((stateg, statec)) return Indexed(nsteptp, stateg, statec) info = TransitionMatrixOf(self, trans_probs) & StochasticStateSpaceOf( self, state_space) new_gc = given_condition & info if isinstance(condition, And): # handle queries like, # P(Eq(X[i+k], s1) & Eq(X[i+m], s2) . . . & Eq(X[i], sn), Eq(P(Eq(X[i], si)), prob)) conds = condition.args idx2state = dict() for cond in conds: idx, state = (cond.lhs, cond.rhs) if isinstance(cond.lhs, RandomIndexedSymbol) else \ (cond.rhs, cond.lhs) idx2state[idx] = cond if idx2state.get(idx, None) is None else \ idx2state[idx] & cond if any( len(Intersection(idx2state[idx].as_set(), state_space)) != 1 for idx in idx2state): return S.Zero # a RandomIndexedSymbol cannot go to different states simultaneously i, result = -1, 1 conds = And.fromiter( Intersection(idx2state[idx].as_set(), state_space).as_relational(idx) for idx in idx2state) if not isinstance(conds, And): return self.probability(conds, new_gc) conds = conds.args while i > -len(conds): result *= self.probability(conds[i], conds[i - 1] & info) i -= 1 if isinstance(given_condition, (TransitionMatrixOf, StochasticStateSpaceOf)): return result * Probability(conds[i]) if isinstance(given_condition, And): idx_sym = conds[i].atoms(RandomIndexedSymbol) prob, count = S(0), 0 for gc in given_condition.args: if gc.atoms(RandomIndexedSymbol) == idx_sym: prob += gc.rhs if isinstance(gc.lhs, Probability) else gc.lhs count += 1 if isinstance(state_space, FiniteSet) and \ count == len(state_space) - 1: given_condition = Eq(Probability(conds[i]), S(1) - prob) if isinstance(given_condition, Eq): if not isinstance(given_condition.lhs, Probability) or \ given_condition.lhs.args[0] != conds[i]: raise ValueError("Probability for %s needed", conds[i]) return result * given_condition.rhs if isinstance(condition, Or): conds, prob_sum = condition.args, S(0) idx2state = dict() for cond in conds: idx, state = (cond.lhs, cond.rhs) if isinstance(cond.lhs, RandomIndexedSymbol) else \ (cond.rhs, cond.lhs) idx2state[idx] = cond if idx2state.get(idx, None) is None else \ idx2state[idx] | cond conds = Or.fromiter( Intersection(idx2state[idx].as_set(), state_space).as_relational(idx) for idx in idx2state) if not isinstance(conds, Or): return self.probability(conds, new_gc) return sum([self.probability(cond, new_gc) for cond in conds.args]) if isinstance(condition, Ne): prob = self.probability(Not(condition), new_gc) return S(1) - prob raise NotImplementedError( "Mechanism for handling (%s, %s) queries hasn't been " "implemented yet." % (condition, given_condition))
def test_DiscreteMarkovChain(): # pass only the name X = DiscreteMarkovChain("X") assert isinstance(X.state_space, Range) assert X.index_set == S.Naturals0 assert isinstance(X.transition_probabilities, MatrixSymbol) t = symbols('t', positive=True, integer=True) assert isinstance(X[t], RandomIndexedSymbol) assert E(X[0]) == Expectation(X[0]) raises(TypeError, lambda: DiscreteMarkovChain(1)) raises(NotImplementedError, lambda: X(t)) nz = Symbol('n', integer=True) TZ = MatrixSymbol('M', nz, nz) SZ = Range(nz) YZ = DiscreteMarkovChain('Y', SZ, TZ) assert P(Eq(YZ[2], 1), Eq(YZ[1], 0)) == TZ[0, 1] raises(ValueError, lambda: sample_stochastic_process(t)) raises(ValueError, lambda: next(sample_stochastic_process(X))) # pass name and state_space # any hashable object should be a valid state # states should be valid as a tuple/set/list/Tuple/Range sym, rainy, cloudy, sunny = symbols('a Rainy Cloudy Sunny', real=True) state_spaces = [(1, 2, 3), [Str('Hello'), sym, DiscreteMarkovChain], Tuple(1, exp(sym), Str('World'), sympify=False), Range(-1, 5, 2), [rainy, cloudy, sunny]] chains = [ DiscreteMarkovChain("Y", state_space) for state_space in state_spaces ] for i, Y in enumerate(chains): assert isinstance(Y.transition_probabilities, MatrixSymbol) assert Y.state_space == state_spaces[i] or Y.state_space == FiniteSet( *state_spaces[i]) assert Y.number_of_states == 3 with ignore_warnings( UserWarning): # TODO: Restore tests once warnings are removed assert P(Eq(Y[2], 1), Eq(Y[0], 2), evaluate=False) == Probability(Eq(Y[2], 1), Eq(Y[0], 2)) assert E(Y[0]) == Expectation(Y[0]) raises(ValueError, lambda: next(sample_stochastic_process(Y))) raises(TypeError, lambda: DiscreteMarkovChain("Y", dict((1, 1)))) Y = DiscreteMarkovChain("Y", Range(1, t, 2)) assert Y.number_of_states == ceiling((t - 1) / 2) # pass name and transition_probabilities chains = [ DiscreteMarkovChain("Y", trans_probs=Matrix([[]])), DiscreteMarkovChain("Y", trans_probs=Matrix([[0, 1], [1, 0]])), DiscreteMarkovChain("Y", trans_probs=Matrix([[pi, 1 - pi], [sym, 1 - sym]])) ] for Z in chains: assert Z.number_of_states == Z.transition_probabilities.shape[0] assert isinstance(Z.transition_probabilities, ImmutableDenseMatrix) # pass name, state_space and transition_probabilities T = Matrix([[0.5, 0.2, 0.3], [0.2, 0.5, 0.3], [0.2, 0.3, 0.5]]) TS = MatrixSymbol('T', 3, 3) Y = DiscreteMarkovChain("Y", [0, 1, 2], T) YS = DiscreteMarkovChain("Y", ['One', 'Two', 3], TS) assert YS._transient2transient() == None assert YS._transient2absorbing() == None assert Y.joint_distribution(1, Y[2], 3) == JointDistribution(Y[1], Y[2], Y[3]) raises(ValueError, lambda: Y.joint_distribution(Y[1].symbol, Y[2].symbol)) assert P(Eq(Y[3], 2), Eq(Y[1], 1)).round(2) == Float(0.36, 2) assert (P(Eq(YS[3], 2), Eq(YS[1], 1)) - (TS[0, 2] * TS[1, 0] + TS[1, 1] * TS[1, 2] + TS[1, 2] * TS[2, 2])).simplify() == 0 assert P(Eq(YS[1], 1), Eq(YS[2], 2)) == Probability(Eq(YS[1], 1)) assert P(Eq(YS[3], 3), Eq( YS[1], 1)) == TS[0, 2] * TS[1, 0] + TS[1, 1] * TS[1, 2] + TS[1, 2] * TS[2, 2] TO = Matrix([[0.25, 0.75, 0], [0, 0.25, 0.75], [0.75, 0, 0.25]]) assert P(Eq(Y[3], 2), Eq(Y[1], 1) & TransitionMatrixOf(Y, TO)).round(3) == Float( 0.375, 3) with ignore_warnings( UserWarning): ### TODO: Restore tests once warnings are removed assert E(Y[3], evaluate=False) == Expectation(Y[3]) assert E(Y[3], Eq(Y[2], 1)).round(2) == Float(1.1, 3) TSO = MatrixSymbol('T', 4, 4) raises( ValueError, lambda: str(P(Eq(YS[3], 2), Eq(YS[1], 1) & TransitionMatrixOf(YS, TSO)))) raises(TypeError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], symbols('M'))) raises( ValueError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], MatrixSymbol('T', 3, 4))) raises(ValueError, lambda: E(Y[3], Eq(Y[2], 6))) raises(ValueError, lambda: E(Y[2], Eq(Y[3], 1))) # extended tests for probability queries TO1 = Matrix([[Rational(1, 4), Rational(3, 4), 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) assert P( And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Probability(Eq(Y[0], 0)), Rational(1, 4)) & TransitionMatrixOf(Y, TO1)) == Rational(1, 16) assert P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), TransitionMatrixOf(Y, TO1)) == \ Probability(Eq(Y[0], 0))/4 assert P( Lt(X[1], 2) & Gt(X[1], 0), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1)) == Rational(1, 4) assert P( Lt(X[1], 2) & Gt(X[1], 0), Eq(X[0], 2) & StochasticStateSpaceOf(X, [None, 'None', 1]) & TransitionMatrixOf(X, TO1)) == Rational(1, 4) assert P( Ne(X[1], 2) & Ne(X[1], 1), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1)) is S.Zero assert P( Ne(X[1], 2) & Ne(X[1], 1), Eq(X[0], 2) & StochasticStateSpaceOf(X, [None, 'None', 1]) & TransitionMatrixOf(X, TO1)) is S.Zero assert P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Y[1], 1)) == 0.1 * Probability(Eq(Y[0], 0)) # testing properties of Markov chain TO2 = Matrix([[S.One, 0, 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) TO3 = Matrix([[Rational(1, 4), Rational(3, 4), 0], [Rational(1, 3), Rational(1, 3), Rational(1, 3)], [0, Rational(1, 4), Rational(3, 4)]]) Y2 = DiscreteMarkovChain('Y', trans_probs=TO2) Y3 = DiscreteMarkovChain('Y', trans_probs=TO3) assert Y3._transient2absorbing() == None raises(ValueError, lambda: Y3.fundamental_matrix()) assert Y2.is_absorbing_chain() == True assert Y3.is_absorbing_chain() == False TO4 = Matrix([[Rational(1, 5), Rational(2, 5), Rational(2, 5)], [Rational(1, 10), S.Half, Rational(2, 5)], [Rational(3, 5), Rational(3, 10), Rational(1, 10)]]) Y4 = DiscreteMarkovChain('Y', trans_probs=TO4) w = ImmutableMatrix([[Rational(11, 39), Rational(16, 39), Rational(4, 13)]]) assert Y4.limiting_distribution == w assert Y4.is_regular() == True TS1 = MatrixSymbol('T', 3, 3) Y5 = DiscreteMarkovChain('Y', trans_probs=TS1) assert Y5.limiting_distribution(w, TO4).doit() == True assert Y5.stationary_distribution(condition_set=True).subs( TS1, TO4).contains(w).doit() == S.true TO6 = Matrix([[S.One, 0, 0, 0, 0], [S.Half, 0, S.Half, 0, 0], [0, S.Half, 0, S.Half, 0], [0, 0, S.Half, 0, S.Half], [0, 0, 0, 0, 1]]) Y6 = DiscreteMarkovChain('Y', trans_probs=TO6) assert Y6._transient2absorbing() == ImmutableMatrix([[S.Half, 0], [0, 0], [0, S.Half]]) assert Y6._transient2transient() == ImmutableMatrix([[0, S.Half, 0], [S.Half, 0, S.Half], [0, S.Half, 0]]) assert Y6.fundamental_matrix() == ImmutableMatrix( [[Rational(3, 2), S.One, S.Half], [S.One, S(2), S.One], [S.Half, S.One, Rational(3, 2)]]) assert Y6.absorbing_probabilities() == ImmutableMatrix( [[Rational(3, 4), Rational(1, 4)], [S.Half, S.Half], [Rational(1, 4), Rational(3, 4)]]) # test for zero-sized matrix functionality X = DiscreteMarkovChain('X', trans_probs=Matrix([[]])) assert X.number_of_states == 0 assert X.stationary_distribution() == Matrix([[]]) # test communication_class # see https://drive.google.com/drive/folders/1HbxLlwwn2b3U8Lj7eb_ASIUb5vYaNIjg?usp=sharing # tutorial 2.pdf TO7 = Matrix([[0, 5, 5, 0, 0], [0, 0, 0, 10, 0], [5, 0, 5, 0, 0], [0, 10, 0, 0, 0], [0, 3, 0, 3, 4]]) / 10 Y7 = DiscreteMarkovChain('Y', trans_probs=TO7) tuples = Y7.communication_classes() classes, recurrence, periods = list(zip(*tuples)) assert classes == ([1, 3], [0, 2], [4]) assert recurrence == (True, False, False) assert periods == (2, 1, 1) TO8 = Matrix([[0, 0, 0, 10, 0, 0], [5, 0, 5, 0, 0, 0], [0, 4, 0, 0, 0, 6], [10, 0, 0, 0, 0, 0], [0, 10, 0, 0, 0, 0], [0, 0, 0, 5, 5, 0] ]) / 10 Y8 = DiscreteMarkovChain('Y', trans_probs=TO8) tuples = Y8.communication_classes() classes, recurrence, periods = list(zip(*tuples)) assert classes == ([0, 3], [1, 2, 5, 4]) assert recurrence == (True, False) assert periods == (2, 2) TO9 = Matrix( [[2, 0, 0, 3, 0, 0, 3, 2, 0, 0], [0, 10, 0, 0, 0, 0, 0, 0, 0, 0], [0, 2, 2, 0, 0, 0, 0, 0, 3, 3], [0, 0, 0, 3, 0, 0, 6, 1, 0, 0], [0, 0, 0, 0, 5, 5, 0, 0, 0, 0], [0, 0, 0, 0, 0, 10, 0, 0, 0, 0], [4, 0, 0, 5, 0, 0, 1, 0, 0, 0], [2, 0, 0, 4, 0, 0, 2, 2, 0, 0], [3, 0, 1, 0, 0, 0, 0, 0, 4, 2], [0, 0, 4, 0, 0, 0, 0, 0, 3, 3]]) / 10 Y9 = DiscreteMarkovChain('Y', trans_probs=TO9) tuples = Y9.communication_classes() classes, recurrence, periods = list(zip(*tuples)) assert classes == ([0, 3, 6, 7], [1], [2, 8, 9], [5], [4]) assert recurrence == (True, True, False, True, False) assert periods == (1, 1, 1, 1, 1) # test custom state space Y10 = DiscreteMarkovChain('Y', [1, 2, 3], TO2) tuples = Y10.communication_classes() classes, recurrence, periods = list(zip(*tuples)) assert classes == ([1], [2, 3]) assert recurrence == (True, False) assert periods == (1, 1) # testing miscellaneous queries T = Matrix([[S.Half, Rational(1, 4), Rational(1, 4)], [Rational(1, 3), 0, Rational(2, 3)], [S.Half, S.Half, 0]]) X = DiscreteMarkovChain('X', [0, 1, 2], T) assert P( Eq(X[1], 2) & Eq(X[2], 1) & Eq(X[3], 0), Eq(P(Eq(X[1], 0)), Rational(1, 4)) & Eq(P(Eq(X[1], 1)), Rational(1, 4))) == Rational(1, 12) assert P(Eq(X[2], 1) | Eq(X[2], 2), Eq(X[1], 1)) == Rational(2, 3) assert P(Eq(X[2], 1) & Eq(X[2], 2), Eq(X[1], 1)) is S.Zero assert P(Ne(X[2], 2), Eq(X[1], 1)) == Rational(1, 3) assert E(X[1]**2, Eq(X[0], 1)) == Rational(8, 3) assert variance(X[1], Eq(X[0], 1)) == Rational(8, 9) raises(ValueError, lambda: E(X[1], Eq(X[2], 1))) raises(ValueError, lambda: DiscreteMarkovChain('X', [0, 1], T)) # testing miscellaneous queries with different state space X = DiscreteMarkovChain('X', ['A', 'B', 'C'], T) assert P( Eq(X[1], 2) & Eq(X[2], 1) & Eq(X[3], 0), Eq(P(Eq(X[1], 0)), Rational(1, 4)) & Eq(P(Eq(X[1], 1)), Rational(1, 4))) == Rational(1, 12) assert P(Eq(X[2], 1) | Eq(X[2], 2), Eq(X[1], 1)) == Rational(2, 3) assert P(Eq(X[2], 1) & Eq(X[2], 2), Eq(X[1], 1)) is S.Zero assert P(Ne(X[2], 2), Eq(X[1], 1)) == Rational(1, 3) a = X.state_space.args[0] c = X.state_space.args[2] assert (E(X[1]**2, Eq(X[0], 1)) - (a**2 / 3 + 2 * c**2 / 3)).simplify() == 0 assert (variance(X[1], Eq(X[0], 1)) - (2 * (-a / 3 + c / 3)**2 / 3 + (2 * a / 3 - 2 * c / 3)**2 / 3)).simplify() == 0 raises(ValueError, lambda: E(X[1], Eq(X[2], 1)))
def test_ContinuousMarkovChain(): T1 = Matrix([[S(-2), S(2), S.Zero], [S.Zero, S.NegativeOne, S.One], [Rational(3, 2), Rational(3, 2), S(-3)]]) C1 = ContinuousMarkovChain('C', [0, 1, 2], T1) assert C1.limiting_distribution() == ImmutableMatrix( [[Rational(3, 19), Rational(12, 19), Rational(4, 19)]]) T2 = Matrix([[-S.One, S.One, S.Zero], [S.One, -S.One, S.Zero], [S.Zero, S.One, -S.One]]) C2 = ContinuousMarkovChain('C', [0, 1, 2], T2) A, t = C2.generator_matrix, symbols('t', positive=True) assert C2.transition_probabilities(A)(t) == Matrix( [[S.Half + exp(-2 * t) / 2, S.Half - exp(-2 * t) / 2, 0], [S.Half - exp(-2 * t) / 2, S.Half + exp(-2 * t) / 2, 0], [ S.Half - exp(-t) + exp(-2 * t) / 2, S.Half - exp(-2 * t) / 2, exp(-t) ]]) with ignore_warnings( UserWarning): ### TODO: Restore tests once warnings are removed assert P(Eq(C2(1), 1), Eq(C2(0), 1), evaluate=False) == Probability(Eq(C2(1), 1), Eq(C2(0), 1)) assert P(Eq(C2(1), 1), Eq(C2(0), 1)) == exp(-2) / 2 + S.Half assert P( Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 1), Eq(P(Eq(C2(1), 0)), S.Half)) == (Rational(1, 4) - exp(-2) / 4) * (exp(-2) / 2 + S.Half) assert P( Not(Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 2)) | (Eq(C2(1), 0) & Eq(C2(2), 1) & Eq(C2(3), 2)), Eq(P(Eq(C2(1), 0)), Rational(1, 4)) & Eq(P(Eq(C2(1), 1)), Rational(1, 4))) is S.One assert E(C2(Rational(3, 2)), Eq(C2(0), 2)) == -exp(-3) / 2 + 2 * exp(Rational(-3, 2)) + S.Half assert variance(C2(Rational(3, 2)), Eq( C2(0), 1)) == ((S.Half - exp(-3) / 2)**2 * (exp(-3) / 2 + S.Half) + (Rational(-1, 2) - exp(-3) / 2)**2 * (S.Half - exp(-3) / 2)) raises(KeyError, lambda: P(Eq(C2(1), 0), Eq(P(Eq(C2(1), 1)), S.Half))) assert P(Eq(C2(1), 0), Eq(P(Eq(C2(5), 1)), S.Half)) == Probability(Eq(C2(1), 0)) TS1 = MatrixSymbol('G', 3, 3) CS1 = ContinuousMarkovChain('C', [0, 1, 2], TS1) A = CS1.generator_matrix assert CS1.transition_probabilities(A)(t) == exp(t * A) C3 = ContinuousMarkovChain( 'C', [Symbol('0'), Symbol('1'), Symbol('2')], T2) assert P(Eq(C3(1), 1), Eq(C3(0), 1)) == exp(-2) / 2 + S.Half assert P(Eq(C3(1), Symbol('1')), Eq(C3(0), Symbol('1'))) == exp(-2) / 2 + S.Half #test probability queries G = Matrix([[-S(1), Rational(1, 10), Rational(9, 10)], [Rational(2, 5), -S(1), Rational(3, 5)], [Rational(1, 2), Rational(1, 2), -S(1)]]) C = ContinuousMarkovChain('C', state_space=[0, 1, 2], gen_mat=G) assert P(Eq(C(7.385), C(3.19)), Eq(C(0.862), 0)).round(5) == Float(0.35469, 5) assert P(Gt(C(98.715), C(19.807)), Eq(C(11.314), 2)).round(5) == Float(0.32452, 5) assert P(Le(C(5.9), C(10.112)), Eq(C(4), 1)).round(6) == Float(0.675214, 6) assert Float(P(Eq(C(7.32), C(2.91)), Eq(C(2.63), 1)), 14) == Float(1 - P(Ne(C(7.32), C(2.91)), Eq(C(2.63), 1)), 14) assert Float(P(Gt(C(3.36), C(1.101)), Eq(C(0.8), 2)), 14) == Float(1 - P(Le(C(3.36), C(1.101)), Eq(C(0.8), 2)), 14) assert Float(P(Lt(C(4.9), C(2.79)), Eq(C(1.61), 0)), 14) == Float(1 - P(Ge(C(4.9), C(2.79)), Eq(C(1.61), 0)), 14) assert P(Eq(C(5.243), C(10.912)), Eq(C(2.174), 1)) == P(Eq(C(10.912), C(5.243)), Eq(C(2.174), 1)) assert P(Gt(C(2.344), C(9.9)), Eq(C(1.102), 1)) == P(Lt(C(9.9), C(2.344)), Eq(C(1.102), 1)) assert P(Ge(C(7.87), C(1.008)), Eq(C(0.153), 1)) == P(Le(C(1.008), C(7.87)), Eq(C(0.153), 1)) #test symbolic queries a, b, c, d = symbols('a b c d') query = P(Eq(C(a), b), Eq(C(c), d)) assert query.subs({ a: 3.65, b: 2, c: 1.78, d: 1 }).evalf().round(10) == P(Eq(C(3.65), 2), Eq(C(1.78), 1)).round(10) query_gt = P(Gt(C(a), b), Eq(C(c), d)) query_le = P(Le(C(a), b), Eq(C(c), d)) assert query_gt.subs({ a: 13.2, b: 0, c: 3.29, d: 2 }).evalf() + query_le.subs({ a: 13.2, b: 0, c: 3.29, d: 2 }).evalf() == 1 query_ge = P(Ge(C(a), b), Eq(C(c), d)) query_lt = P(Lt(C(a), b), Eq(C(c), d)) assert query_ge.subs({ a: 7.43, b: 1, c: 1.45, d: 0 }).evalf() + query_lt.subs({ a: 7.43, b: 1, c: 1.45, d: 0 }).evalf() == 1 #test issue 20078 assert (2 * C(1) + 3 * C(1)).simplify() == 5 * C(1) assert (2 * C(1) - 3 * C(1)).simplify() == -C(1) assert (2 * (0.25 * C(1))).simplify() == 0.5 * C(1) assert (2 * C(1) * 0.25 * C(1)).simplify() == 0.5 * C(1)**2 assert (C(1)**2 + C(1)**3).simplify() == (C(1) + 1) * C(1)**2
def test_DiscreteMarkovChain(): # pass only the name X = DiscreteMarkovChain("X") assert X.state_space == S.Reals assert X.index_set == S.Naturals0 assert X.transition_probabilities == None t = symbols('t', positive=True, integer=True) assert isinstance(X[t], RandomIndexedSymbol) assert E(X[0]) == Expectation(X[0]) raises(TypeError, lambda: DiscreteMarkovChain(1)) raises(NotImplementedError, lambda: X(t)) # pass name and state_space Y = DiscreteMarkovChain("Y", [1, 2, 3]) assert Y.transition_probabilities == None assert Y.state_space == FiniteSet(1, 2, 3) assert P(Eq(Y[2], 1), Eq(Y[0], 2)) == Probability(Eq(Y[2], 1), Eq(Y[0], 2)) assert E(X[0]) == Expectation(X[0]) raises(TypeError, lambda: DiscreteMarkovChain("Y", dict((1, 1)))) # pass name, state_space and transition_probabilities T = Matrix([[0.5, 0.2, 0.3], [0.2, 0.5, 0.3], [0.2, 0.3, 0.5]]) TS = MatrixSymbol('T', 3, 3) Y = DiscreteMarkovChain("Y", [0, 1, 2], T) YS = DiscreteMarkovChain("Y", [0, 1, 2], TS) assert YS._transient2transient() == None assert YS._transient2absorbing() == None assert Y.joint_distribution(1, Y[2], 3) == JointDistribution(Y[1], Y[2], Y[3]) raises(ValueError, lambda: Y.joint_distribution(Y[1].symbol, Y[2].symbol)) assert P(Eq(Y[3], 2), Eq(Y[1], 1)).round(2) == Float(0.36, 2) assert str(P(Eq(YS[3], 2), Eq(YS[1], 1))) == \ "T[0, 2]*T[1, 0] + T[1, 1]*T[1, 2] + T[1, 2]*T[2, 2]" TO = Matrix([[0.25, 0.75, 0], [0, 0.25, 0.75], [0.75, 0, 0.25]]) assert P(Eq(Y[3], 2), Eq(Y[1], 1) & TransitionMatrixOf(Y, TO)).round(3) == Float( 0.375, 3) assert E(Y[3], evaluate=False) == Expectation(Y[3]) assert E(Y[3], Eq(Y[2], 1)).round(2) == Float(1.1, 3) TSO = MatrixSymbol('T', 4, 4) raises( ValueError, lambda: str(P(Eq(YS[3], 2), Eq(YS[1], 1) & TransitionMatrixOf(YS, TSO)))) raises(TypeError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], symbols('M'))) raises( ValueError, lambda: DiscreteMarkovChain("Z", [0, 1, 2], MatrixSymbol('T', 3, 4))) raises(IndexError, lambda: str(P(Eq(YS[3], 3), Eq(YS[1], 1)))) raises(ValueError, lambda: str(P(Eq(YS[1], 1), Eq(YS[2], 2)))) raises(ValueError, lambda: E(Y[3], Eq(Y[2], 6))) raises(ValueError, lambda: E(Y[2], Eq(Y[3], 1))) # extended tests for probability queries TO1 = Matrix([[S(1) / 4, S(3) / 4, 0], [S(1) / 3, S(1) / 3, S(1) / 3], [0, S(1) / 4, S(3) / 4]]) assert P( And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Probability(Eq(Y[0], 0)), S(1) / 4) & TransitionMatrixOf(Y, TO1)) == S(1) / 16 assert P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), TransitionMatrixOf(Y, TO1)) == \ Probability(Eq(Y[0], 0))/4 assert P( Lt(X[1], 2) & Gt(X[1], 0), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1)) == S(1) / 4 assert P( Ne(X[1], 2) & Ne(X[1], 1), Eq(X[0], 2) & StochasticStateSpaceOf(X, [0, 1, 2]) & TransitionMatrixOf(X, TO1)) == S(0) raises( ValueError, lambda: str( P(And(Eq(Y[2], 1), Eq(Y[1], 1), Eq(Y[0], 0)), Eq(Y[1], 1)))) # testing properties of Markov chain TO2 = Matrix([[S(1), 0, 0], [S(1) / 3, S(1) / 3, S(1) / 3], [0, S(1) / 4, S(3) / 4]]) TO3 = Matrix([[S(1) / 4, S(3) / 4, 0], [S(1) / 3, S(1) / 3, S(1) / 3], [0, S(1) / 4, S(3) / 4]]) Y2 = DiscreteMarkovChain('Y', trans_probs=TO2) Y3 = DiscreteMarkovChain('Y', trans_probs=TO3) assert Y3._transient2absorbing() == None raises(ValueError, lambda: Y3.fundamental_matrix()) assert Y2.is_absorbing_chain() == True assert Y3.is_absorbing_chain() == False TO4 = Matrix([[S(1) / 5, S(2) / 5, S(2) / 5], [S(1) / 10, S(1) / 2, S(2) / 5], [S(3) / 5, S(3) / 10, S(1) / 10]]) Y4 = DiscreteMarkovChain('Y', trans_probs=TO4) w = ImmutableMatrix([[S(11) / 39, S(16) / 39, S(4) / 13]]) assert Y4.limiting_distribution == w assert Y4.is_regular() == True TS1 = MatrixSymbol('T', 3, 3) Y5 = DiscreteMarkovChain('Y', trans_probs=TS1) assert Y5.limiting_distribution(w, TO4).doit() == True TO6 = Matrix([[S(1), 0, 0, 0, 0], [S(1) / 2, 0, S(1) / 2, 0, 0], [0, S(1) / 2, 0, S(1) / 2, 0], [0, 0, S(1) / 2, 0, S(1) / 2], [0, 0, 0, 0, 1]]) Y6 = DiscreteMarkovChain('Y', trans_probs=TO6) assert Y6._transient2absorbing() == ImmutableMatrix([[S(1) / 2, 0], [0, 0], [0, S(1) / 2]]) assert Y6._transient2transient() == ImmutableMatrix( [[0, S(1) / 2, 0], [S(1) / 2, 0, S(1) / 2], [0, S(1) / 2, 0]]) assert Y6.fundamental_matrix() == ImmutableMatrix( [[S(3) / 2, S(1), S(1) / 2], [S(1), S(2), S(1)], [S(1) / 2, S(1), S(3) / 2]]) assert Y6.absorbing_probabilites() == ImmutableMatrix( [[S(3) / 4, S(1) / 4], [S(1) / 2, S(1) / 2], [S(1) / 4, S(3) / 4]]) # testing miscellaneous queries T = Matrix([[S(1) / 2, S(1) / 4, S(1) / 4], [S(1) / 3, 0, S(2) / 3], [S(1) / 2, S(1) / 2, 0]]) X = DiscreteMarkovChain('X', [0, 1, 2], T) assert P( Eq(X[1], 2) & Eq(X[2], 1) & Eq(X[3], 0), Eq(P(Eq(X[1], 0)), S(1) / 4) & Eq(P(Eq(X[1], 1)), S(1) / 4)) == S(1) / 12 assert P(Eq(X[2], 1) | Eq(X[2], 2), Eq(X[1], 1)) == S(2) / 3 assert P(Eq(X[2], 1) & Eq(X[2], 2), Eq(X[1], 1)) == S(0) assert P(Ne(X[2], 2), Eq(X[1], 1)) == S(1) / 3 assert E(X[1]**2, Eq(X[0], 1)) == S(8) / 3 assert variance(X[1], Eq(X[0], 1)) == S(8) / 9 raises(ValueError, lambda: E(X[1], Eq(X[2], 1)))
def test_PoissonProcess(): X = PoissonProcess("X", 3) assert X.state_space == S.Naturals0 assert X.index_set == Interval(0, oo) assert X.lamda == 3 t, d, x, y = symbols('t d x y', positive=True) assert isinstance(X(t), RandomIndexedSymbol) assert X.distribution(t) == PoissonDistribution(3 * t) raises(ValueError, lambda: PoissonProcess("X", -1)) raises(NotImplementedError, lambda: X[t]) raises(IndexError, lambda: X(-5)) assert X.joint_distribution(X(2), X(3)) == JointDistributionHandmade( Lambda((X(2), X(3)), 6**X(2) * 9**X(3) * exp(-15) / (factorial(X(2)) * factorial(X(3))))) assert X.joint_distribution(4, 6) == JointDistributionHandmade( Lambda((X(4), X(6)), 12**X(4) * 18**X(6) * exp(-30) / (factorial(X(4)) * factorial(X(6))))) assert P(X(t) < 1) == exp(-3 * t) assert P(Eq(X(t), 0), Contains(t, Interval.Lopen(3, 5))) == exp(-6) # exp(-2*lamda) res = P(Eq(X(t), 1), Contains(t, Interval.Lopen(3, 4))) assert res == 3 * exp(-3) # Equivalent to P(Eq(X(t), 1))**4 because of non-overlapping intervals assert P( Eq(X(t), 1) & Eq(X(d), 1) & Eq(X(x), 1) & Eq(X(y), 1), Contains(t, Interval.Lopen(0, 1)) & Contains(d, Interval.Lopen(1, 2)) & Contains(x, Interval.Lopen(2, 3)) & Contains(y, Interval.Lopen(3, 4))) == res**4 # Return Probability because of overlapping intervals assert P(Eq(X(t), 2) & Eq(X(d), 3), Contains(t, Interval.Lopen(0, 2)) & Contains(d, Interval.Ropen(2, 4))) == \ Probability(Eq(X(d), 3) & Eq(X(t), 2), Contains(t, Interval.Lopen(0, 2)) & Contains(d, Interval.Ropen(2, 4))) raises(ValueError, lambda: P( Eq(X(t), 2) & Eq(X(d), 3), Contains(t, Interval.Lopen(0, 4)) & Contains(d, Interval.Lopen(3, oo))) ) # no bound on d assert P(Eq(X(3), 2)) == 81 * exp(-9) / 2 assert P(Eq(X(t), 2), Contains(t, Interval.Lopen(0, 5))) == 225 * exp(-15) / 2 # Check that probability works correctly by adding it to 1 res1 = P(X(t) <= 3, Contains(t, Interval.Lopen(0, 5))) res2 = P(X(t) > 3, Contains(t, Interval.Lopen(0, 5))) assert res1 == 691 * exp(-15) assert (res1 + res2).simplify() == 1 # Check Not and Or assert P(Not(Eq(X(t), 2) & (X(d) > 3)), Contains(t, Interval.Ropen(2, 4)) & \ Contains(d, Interval.Lopen(7, 8))).simplify() == -18*exp(-6) + 234*exp(-9) + 1 assert P(Eq(X(t), 2) | Ne(X(t), 4), Contains(t, Interval.Ropen(2, 4))) == 1 - 36 * exp(-6) raises(ValueError, lambda: P(X(t) > 2, X(t) + X(d))) assert E( X(t)) == 3 * t # property of the distribution at a given timestamp assert E( X(t)**2 + X(d) * 2 + X(y)**3, Contains(t, Interval.Lopen(0, 1)) & Contains(d, Interval.Lopen(1, 2)) & Contains(y, Interval.Ropen(3, 4))) == 75 assert E(X(t)**2, Contains(t, Interval.Lopen(0, 1))) == 12 assert E(x*(X(t) + X(d))*(X(t)**2+X(d)**2), Contains(t, Interval.Lopen(0, 1)) & Contains(d, Interval.Ropen(1, 2))) == \ Expectation(x*(X(d) + X(t))*(X(d)**2 + X(t)**2), Contains(t, Interval.Lopen(0, 1)) & Contains(d, Interval.Ropen(1, 2))) # Value Error because of infinite time bound raises(ValueError, lambda: E(X(t)**3, Contains(t, Interval.Lopen(1, oo)))) # Equivalent to E(X(t)**2) - E(X(d)**2) == E(X(1)**2) - E(X(1)**2) == 0 assert E((X(t) + X(d)) * (X(t) - X(d)), Contains(t, Interval.Lopen(0, 1)) & Contains(d, Interval.Lopen(1, 2))) == 0 assert E(X(2) + x * E(X(5))) == 15 * x + 6 assert E(x * X(1) + y) == 3 * x + y assert P(Eq(X(1), 2) & Eq(X(t), 3), Contains(t, Interval.Lopen(1, 2))) == 81 * exp(-6) / 4 Y = PoissonProcess("Y", 6) Z = X + Y assert Z.lamda == X.lamda + Y.lamda == 9 raises(ValueError, lambda: X + 5) # should be added be only PoissonProcess instance N, M = Z.split(4, 5) assert N.lamda == 4 assert M.lamda == 5 raises(ValueError, lambda: Z.split(3, 2)) # 2+3 != 9 raises( ValueError, lambda: P(Eq(X(t), 0), Contains(t, Interval.Lopen(1, 3)) & Eq(X(1), 0))) # check if it handles queries with two random variables in one args res1 = P(Eq(N(3), N(5))) assert res1 == P(Eq(N(t), 0), Contains(t, Interval(3, 5))) res2 = P(N(3) > N(1)) assert res2 == P((N(t) > 0), Contains(t, Interval(1, 3))) assert P(N(3) < N(1)) == 0 # condition is not possible res3 = P(N(3) <= N(1)) # holds only for Eq(N(3), N(1)) assert res3 == P(Eq(N(t), 0), Contains(t, Interval(1, 3))) # tests from https://www.probabilitycourse.com/chapter11/11_1_2_basic_concepts_of_the_poisson_process.php X = PoissonProcess('X', 10) # 11.1 assert P(Eq(X(S(1) / 3), 3) & Eq(X(1), 10)) == exp(-10) * Rational(8000000000, 11160261) assert P(Eq(X(1), 1), Eq(X(S(1) / 3), 3)) == 0 assert P(Eq(X(1), 10), Eq(X(S(1) / 3), 3)) == P(Eq(X(S(2) / 3), 7)) X = PoissonProcess('X', 2) # 11.2 assert P(X(S(1) / 2) < 1) == exp(-1) assert P(X(3) < 1, Eq(X(1), 0)) == exp(-4) assert P(Eq(X(4), 3), Eq(X(2), 3)) == exp(-4) X = PoissonProcess('X', 3) assert P(Eq(X(2), 5) & Eq(X(1), 2)) == Rational(81, 4) * exp(-6) # check few properties assert P( X(2) <= 3, X(1) >= 1) == 3 * P(Eq(X(1), 0)) + 2 * P(Eq(X(1), 1)) + P(Eq(X(1), 2)) assert P(X(2) <= 3, X(1) > 1) == 2 * P(Eq(X(1), 0)) + 1 * P(Eq(X(1), 1)) assert P(Eq(X(2), 5) & Eq(X(1), 2)) == P(Eq(X(1), 3)) * P(Eq(X(1), 2)) assert P(Eq(X(3), 4), Eq(X(1), 3)) == P(Eq(X(2), 1)) #test issue 20078 assert (2 * X(t) + 3 * X(t)).simplify() == 5 * X(t) assert (2 * X(t) - 3 * X(t)).simplify() == -X(t) assert (2 * (0.25 * X(t))).simplify() == 0.5 * X(t) assert (2 * X(t) * 0.25 * X(t)).simplify() == 0.5 * X(t)**2 assert (X(t)**2 + X(t)**3).simplify() == (X(t) + 1) * X(t)**2
def probability(self, condition, given_condition=None, evaluate=True, **kwargs): """ Handles probability queries for discrete Markov chains. Parameters ========== condition: Relational given_condition: Relational/And Returns ======= Probability If the transition probabilities are not available Expr If the transition probabilities is MatrixSymbol or Matrix 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 and state space using StochasticStateSpaceOf in given_condition using & or And. """ check, trans_probs, state_space, given_condition = \ self._preprocess(given_condition, evaluate) if check: return Probability(condition, given_condition) if isinstance(condition, Eq) and \ isinstance(given_condition, Eq) and \ len(given_condition.atoms(RandomSymbol)) == 1: # handles simple queries like P(Eq(X[i], dest_state), Eq(X[i], init_state)) lhsc, rhsc = condition.lhs, condition.rhs lhsg, rhsg = given_condition.lhs, given_condition.rhs if not isinstance(lhsc, RandomIndexedSymbol): lhsc, rhsc = (rhsc, lhsc) if not isinstance(lhsg, RandomIndexedSymbol): lhsg, rhsg = (rhsg, lhsg) keyc, statec, keyg, stateg = (lhsc.key, rhsc, lhsg.key, rhsg) if Lt(stateg, trans_probs.shape[0]) == False or Lt( statec, trans_probs.shape[1]) == False: raise IndexError( "No information is avaliable for (%s, %s) in " "transition probabilities of shape, (%s, %s). " "State space is zero indexed." % (stateg, statec, trans_probs.shape[0], trans_probs.shape[1])) if keyc < keyg: raise ValueError( "Incorrect given condition is given, probability " "of past state cannot be computed from future state.") nsteptp = trans_probs**(keyc - keyg) if hasattr(nsteptp, "__getitem__"): return nsteptp.__getitem__((stateg, statec)) return Indexed(nsteptp, stateg, statec) if isinstance(condition, And): # handle queries like, # P(Eq(X[i+k], s1) & Eq(X[i+m], s2) . . . & Eq(X[i], sn), Eq(P(X[i]), prob)) conds = condition.args i, result = -1, 1 while i > -len(conds): result *= self.probability(conds[i], conds[i-1] & \ TransitionMatrixOf(self, trans_probs) & \ StochasticStateSpaceOf(self, state_space)) i -= 1 if isinstance(given_condition, (TransitionMatrixOf, StochasticStateSpaceOf)): return result * Probability(conds[i]) if isinstance(given_condition, Eq): if not isinstance(given_condition.lhs, Probability) or \ given_condition.lhs.args[0] != conds[i]: raise ValueError("Probability for %s needed", conds[i]) return result * given_condition.rhs raise NotImplementedError( "Mechanism for handling (%s, %s) queries hasn't been " "implemented yet." % (condition, given_condition))