From 8b6b15a8a1f1fe7bb0e0f91e82f9e3aaf7e8319c Mon Sep 17 00:00:00 2001
From: Chandan Kumar Roy <chandanroy559@outlook.com>
Date: Tue, 29 Dec 2020 18:13:11 +0530
Subject: [PATCH 1/7] asymmetric laplace distribution added
---
pymc3/distributions/__init__.py | 2 ++
pymc3/distributions/continuous.py | 53 +++++++++++++++++++++++++++++++
pymc3/tests/test_distributions.py | 17 ++++++++--
3 files changed, 70 insertions(+), 2 deletions(-)
diff --git a/pymc3/distributions/__init__.py b/pymc3/distributions/__init__.py
index ade7c9ef00..afbdf76b04 100644
--- a/pymc3/distributions/__init__.py
+++ b/pymc3/distributions/__init__.py
@@ -16,6 +16,7 @@
from pymc3.distributions.bart import BART
from pymc3.distributions.bound import Bound
from pymc3.distributions.continuous import (
+ AsymmetricLaplace,
Beta,
Cauchy,
ChiSquared,
@@ -160,6 +161,7 @@
"LKJCorr",
"AR1",
"AR",
+ "AsymmetricLaplace",
"GaussianRandomWalk",
"MvGaussianRandomWalk",
"MvStudentTRandomWalk",
diff --git a/pymc3/distributions/continuous.py b/pymc3/distributions/continuous.py
index 25a97f7688..402cfe9ffa 100644
--- a/pymc3/distributions/continuous.py
+++ b/pymc3/distributions/continuous.py
@@ -80,6 +80,7 @@
"Interpolated",
"Rice",
"Moyal",
+ "AsymmetricLaplace",
]
@@ -1660,6 +1661,58 @@ def logcdf(self, value):
tt.switch(tt.gt(y, 1), tt.log1p(-0.5 * tt.exp(-y)), tt.log(1 - 0.5 * tt.exp(-y))),
)
+class AsymmetricLaplace(Continuous):
+ """
+ Assymetric Laplace Distribution
+ See also: https://en.wikipedia.org/wiki/Asymmetric_Laplace_distribution
+ """
+
+ def __init__(self, scale, symmetry, testval=0.0, *args, **kwargs):
+ """
+ Constructor
+ :param scale: scale parameter
+ :param symmetry: asymmetry parameter. Reduces to a normal laplace distribution with value 1
+ """
+ self.scale = tt.as_tensor_variable(scale)
+ self.symmetry = tt.as_tensor_variable(symmetry)
+
+ super().__init__(*args, **kwargs, testval=testval)
+
+ def random(self, point=None, size=None):
+ """
+ Draw random samples from this distribution, using the inverse CDF method.
+ :param point: not used
+ :param size: size of sample to draw
+ :return: Samples
+ """
+ if point is not None:
+ raise NotImplementedError('Random not implemented with point specified')
+
+ if size is not None:
+ u = np.random.uniform(size=size)
+ x = - tt.log((1 - u) * (1 + self.symmetry ** 2)) / (self.symmetry * self.scale) * (
+ u > ((self.symmetry ** 2) / (1 + self.symmetry ** 2))) + self.symmetry * tt.log(
+ u * (1 + self.symmetry ** 2) / (self.symmetry ** 2)) / self.scale * (
+ u < ((self.symmetry ** 2) / (1 + self.symmetry ** 2)))
+
+ return x
+
+ u = np.random.uniform()
+ if u > (self.symmetry ** 2) / (1 + self.symmetry ** 2):
+ x = - tt.log((1 - u) * (1 + self.symmetry ** 2)) / (self.symmetry * self.scale)
+ else:
+ x = self.symmetry * tt.log(u * (1 + self.symmetry ** 2) / (self.symmetry ** 2)) / self.scale
+
+ return x
+
+ def logp(self, value):
+ """
+ Compute logp.
+ :param value: evaluation point
+ :return: log probability at evaluation point
+ """
+ return tt.log(self.scale / (self.symmetry + (self.symmetry ** -1))) + (
+ -value * self.scale * tt.sgn(value) * (self.symmetry ** tt.sgn(value)))
class Lognormal(PositiveContinuous):
r"""
diff --git a/pymc3/tests/test_distributions.py b/pymc3/tests/test_distributions.py
index a2d00a2d60..41338f894c 100644
--- a/pymc3/tests/test_distributions.py
+++ b/pymc3/tests/test_distributions.py
@@ -33,6 +33,7 @@
from pymc3.blocking import DictToVarBijection
from pymc3.distributions import (
AR1,
+ AsymmetricLaplace,
Bernoulli,
Beta,
BetaBinomial,
@@ -206,6 +207,7 @@ def product(domains, n_samples=-1):
Bool = Domain([0, 0, 1, 1], "int64")
+Scale = Domain([0.5,1,1.5])
def build_model(distfam, valuedomain, vardomains, extra_args=None):
if extra_args is None:
@@ -218,7 +220,12 @@ def build_model(distfam, valuedomain, vardomains, extra_args=None):
distfam("value", shape=valuedomain.shape, transform=None, **vals)
return m
-
+def laplace_asymmetric_logpdf(value, symmetry,scale = None):
+ kapinv = 1/symmetry
+ lPx = value * np.where(value >= 0, -symmetry, kapinv)
+ lPx -=np.log((symmetry+kapinv))
+ return lPx
+
def integrate_nd(f, domain, shape, dtype):
if shape == () or shape == (1,):
if dtype in continuous_types:
@@ -986,7 +993,13 @@ def test_laplace(self):
{"mu": R, "b": Rplus},
lambda value, mu, b: sp.laplace.logcdf(value, mu, b),
)
-
+ def test_laplace_asymmetric(self):
+ self.pymc3_matches_scipy(
+ AsymmetricLaplace,
+ R,
+ {"scale":Scale,"symmetry":Rplus},
+ laplace_asymmetric_logpdf,
+ )
def test_lognormal(self):
self.pymc3_matches_scipy(
Lognormal,
From b9c335e0c6b8fd0672e469cdcd1a2964376b3c9a Mon Sep 17 00:00:00 2001
From: Chandan Kumar Roy <chandanroy559@outlook.com>
Date: Thu, 31 Dec 2020 23:50:31 +0530
Subject: [PATCH 2/7] unit test for random function added
---
pymc3/distributions/continuous.py | 92 +++++++++++++++---------
pymc3/tests/test_distributions.py | 19 ++---
pymc3/tests/test_distributions_random.py | 20 ++++++
3 files changed, 90 insertions(+), 41 deletions(-)
diff --git a/pymc3/distributions/continuous.py b/pymc3/distributions/continuous.py
index 402cfe9ffa..f808fcd3f0 100644
--- a/pymc3/distributions/continuous.py
+++ b/pymc3/distributions/continuous.py
@@ -1661,58 +1661,84 @@ def logcdf(self, value):
tt.switch(tt.gt(y, 1), tt.log1p(-0.5 * tt.exp(-y)), tt.log(1 - 0.5 * tt.exp(-y))),
)
+
class AsymmetricLaplace(Continuous):
- """
- Assymetric Laplace Distribution
+ r"""
+ Asymmetric-Laplace log-likelihood.
+
+ The pdf of this distribution is
+
+ ..math::
+ {f(x|\\b,\kappa) =
+ \left({\frac{\\b}{\kappa + 1/\kappa}}\right)\,e^{-(x)\\b\,s\kappa ^{s}}}
+ where s = sgn(x)
+
+
See also: https://en.wikipedia.org/wiki/Asymmetric_Laplace_distribution
"""
- def __init__(self, scale, symmetry, testval=0.0, *args, **kwargs):
- """
- Constructor
- :param scale: scale parameter
- :param symmetry: asymmetry parameter. Reduces to a normal laplace distribution with value 1
- """
- self.scale = tt.as_tensor_variable(scale)
- self.symmetry = tt.as_tensor_variable(symmetry)
+ def __init__(self, b, kappa, testval=0.0, *args, **kwargs):
+ self.b = tt.as_tensor_variable(b)
+ self.kappa = tt.as_tensor_variable(kappa)
super().__init__(*args, **kwargs, testval=testval)
- def random(self, point=None, size=None):
- """
- Draw random samples from this distribution, using the inverse CDF method.
- :param point: not used
- :param size: size of sample to draw
- :return: Samples
- """
- if point is not None:
- raise NotImplementedError('Random not implemented with point specified')
-
+ def _random(self, b, kappa, size=None):
if size is not None:
u = np.random.uniform(size=size)
- x = - tt.log((1 - u) * (1 + self.symmetry ** 2)) / (self.symmetry * self.scale) * (
- u > ((self.symmetry ** 2) / (1 + self.symmetry ** 2))) + self.symmetry * tt.log(
- u * (1 + self.symmetry ** 2) / (self.symmetry ** 2)) / self.scale * (
- u < ((self.symmetry ** 2) / (1 + self.symmetry ** 2)))
-
+ x = -np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b) * (
+ u > ((kappa ** 2) / (1 + kappa ** 2))
+ ) + kappa * np.log(u * (1 + kappa ** 2) / (kappa ** 2)) / b * (
+ u < ((kappa ** 2) / (1 + kappa ** 2))
+ )
return x
u = np.random.uniform()
- if u > (self.symmetry ** 2) / (1 + self.symmetry ** 2):
- x = - tt.log((1 - u) * (1 + self.symmetry ** 2)) / (self.symmetry * self.scale)
+ if u > (kappa ** 2) / (1 + kappa ** 2):
+ x = -np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b)
else:
- x = self.symmetry * tt.log(u * (1 + self.symmetry ** 2) / (self.symmetry ** 2)) / self.scale
+ x = kappa * np.log(u * (1 + kappa ** 2) / (kappa ** 2)) / b
return x
+ def random(self, point=None, size=None):
+ """
+ Draw random samples from this distribution, using the inverse CDF method.
+
+ Parameters
+ ----------
+ point: dict, optional
+ Dict of variable values on which random values are to be
+ conditioned (uses default point if not specified).
+ size:int, optional
+ Desired size of random sample (returns one sample if not
+ specified).
+
+ Returns
+ -------
+ array
+ """
+ b, kappa = draw_values([self.b, self.kappa], point=point, size=size)
+ return generate_samples(self._random, b, kappa, dist_shape=self.shape, size=size)
+
def logp(self, value):
"""
- Compute logp.
- :param value: evaluation point
- :return: log probability at evaluation point
+ Calculate log-probability of Asymmetric-Laplace distribution at specified value.
+
+ Parameters
+ ----------
+ value: numeric
+ Value(s) for which log-probability is calculated. If the log probabilities for multiple
+ values are desired the values must be provided in a numpy array or theano tensor
+
+ Returns
+ -------
+ TensorVariable
"""
- return tt.log(self.scale / (self.symmetry + (self.symmetry ** -1))) + (
- -value * self.scale * tt.sgn(value) * (self.symmetry ** tt.sgn(value)))
+ return tt.log(self.b / (self.kappa + (self.kappa ** -1))) + (
+ -value * self.b * tt.sgn(value) * (self.kappa ** tt.sgn(value))
+ )
+
class Lognormal(PositiveContinuous):
r"""
diff --git a/pymc3/tests/test_distributions.py b/pymc3/tests/test_distributions.py
index 41338f894c..948041373e 100644
--- a/pymc3/tests/test_distributions.py
+++ b/pymc3/tests/test_distributions.py
@@ -207,7 +207,6 @@ def product(domains, n_samples=-1):
Bool = Domain([0, 0, 1, 1], "int64")
-Scale = Domain([0.5,1,1.5])
def build_model(distfam, valuedomain, vardomains, extra_args=None):
if extra_args is None:
@@ -220,12 +219,14 @@ def build_model(distfam, valuedomain, vardomains, extra_args=None):
distfam("value", shape=valuedomain.shape, transform=None, **vals)
return m
-def laplace_asymmetric_logpdf(value, symmetry,scale = None):
- kapinv = 1/symmetry
- lPx = value * np.where(value >= 0, -symmetry, kapinv)
- lPx -=np.log((symmetry+kapinv))
+
+def laplace_asymmetric_logpdf(value, kappa, b=None):
+ kapinv = 1 / kappa
+ lPx = value * np.where(value >= 0, -kappa, kapinv)
+ lPx -= np.log(kappa + kapinv)
return lPx
-
+
+
def integrate_nd(f, domain, shape, dtype):
if shape == () or shape == (1,):
if dtype in continuous_types:
@@ -993,13 +994,15 @@ def test_laplace(self):
{"mu": R, "b": Rplus},
lambda value, mu, b: sp.laplace.logcdf(value, mu, b),
)
+
def test_laplace_asymmetric(self):
self.pymc3_matches_scipy(
AsymmetricLaplace,
R,
- {"scale":Scale,"symmetry":Rplus},
+ {"b": Domain([0, 1, inf]), "kappa": Rplus},
laplace_asymmetric_logpdf,
- )
+ )
+
def test_lognormal(self):
self.pymc3_matches_scipy(
Lognormal,
diff --git a/pymc3/tests/test_distributions_random.py b/pymc3/tests/test_distributions_random.py
index 5e735b6aed..eb359091aa 100644
--- a/pymc3/tests/test_distributions_random.py
+++ b/pymc3/tests/test_distributions_random.py
@@ -375,6 +375,11 @@ class TestLaplace(BaseTestCases.BaseTestCase):
params = {"mu": 1.0, "b": 1.0}
+class TestAsymmetricLaplace(BaseTestCases.BaseTestCase):
+ distribution = pm.AsymmetricLaplace
+ params = {"kappa": 1.0, "b": 1.0}
+
+
class TestLognormal(BaseTestCases.BaseTestCase):
distribution = pm.Lognormal
params = {"mu": 1.0, "tau": 1.0}
@@ -626,6 +631,21 @@ def ref_rand(size, mu, b):
pymc3_random(pm.Laplace, {"mu": R, "b": Rplus}, ref_rand=ref_rand)
+ def test_laplace_asymmetric(self):
+ def ref_rand(size, kappa, b):
+ u = np.random.uniform(size=size)
+ x = -np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b) * (
+ u > ((kappa ** 2) / (1 + kappa ** 2))
+ ) + kappa * np.log(u * (1 + kappa ** 2) / (kappa ** 2)) / b * (
+ u < ((kappa ** 2) / (1 + kappa ** 2))
+ )
+
+ return x
+
+ pymc3_random(
+ pm.AsymmetricLaplace, {"b": Domain([0, 1, np.inf]), "kappa": Rplus}, ref_rand=ref_rand
+ )
+
def test_lognormal(self):
def ref_rand(size, mu, tau):
return np.exp(mu + (tau ** -0.5) * st.norm.rvs(loc=0.0, scale=1.0, size=size))
From a6cc0ba824614a1eaf3340d813c4d847684e167a Mon Sep 17 00:00:00 2001
From: Chandan Kumar Roy <chandanroy559@outlook.com>
Date: Sat, 2 Jan 2021 18:54:23 +0530
Subject: [PATCH 3/7] continuous.py updated
---
pymc3/distributions/continuous.py | 13 +++++++++++--
1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/pymc3/distributions/continuous.py b/pymc3/distributions/continuous.py
index f808fcd3f0..1c68287cb6 100644
--- a/pymc3/distributions/continuous.py
+++ b/pymc3/distributions/continuous.py
@@ -1673,6 +1673,12 @@ class AsymmetricLaplace(Continuous):
\left({\frac{\\b}{\kappa + 1/\kappa}}\right)\,e^{-(x)\\b\,s\kappa ^{s}}}
where s = sgn(x)
+ Parameters
+ ----------
+ b:
+ Scale parameter (b > 0)
+ kappa:
+ Symmetry parameter (kappa > 0)
See also: https://en.wikipedia.org/wiki/Asymmetric_Laplace_distribution
"""
@@ -1735,8 +1741,11 @@ def logp(self, value):
-------
TensorVariable
"""
- return tt.log(self.b / (self.kappa + (self.kappa ** -1))) + (
- -value * self.b * tt.sgn(value) * (self.kappa ** tt.sgn(value))
+ return bound(
+ tt.log(self.b / (self.kappa + (self.kappa ** -1)))
+ + (-value * self.b * tt.sgn(value) * (self.kappa ** tt.sgn(value))),
+ 0 < self.b,
+ 0 < self.kappa,
)
From 8b65c66173d3c79caed11e18e55e608616917efb Mon Sep 17 00:00:00 2001
From: Chandan Kumar Roy <chandanroy559@outlook.com>
Date: Tue, 5 Jan 2021 01:18:00 +0530
Subject: [PATCH 4/7] suggested changes added
---
pymc3/distributions/continuous.py | 60 ++++++++++++++----------
pymc3/tests/test_distributions.py | 8 ++--
pymc3/tests/test_distributions_random.py | 4 +-
3 files changed, 40 insertions(+), 32 deletions(-)
diff --git a/pymc3/distributions/continuous.py b/pymc3/distributions/continuous.py
index 1c68287cb6..88cff6779f 100644
--- a/pymc3/distributions/continuous.py
+++ b/pymc3/distributions/continuous.py
@@ -1668,44 +1668,53 @@ class AsymmetricLaplace(Continuous):
The pdf of this distribution is
- ..math::
+ .. math::
{f(x|\\b,\kappa) =
\left({\frac{\\b}{\kappa + 1/\kappa}}\right)\,e^{-(x)\\b\,s\kappa ^{s}}}
- where s = sgn(x)
+
+ where
+
+ .. math::
+
+ s = sgn(x)
+
+ ======== ========================
+ Support :math:`x \in \mathbb{R}`
+ Mean :math:`\mu-\frac{\\\kappa-1/\kappa}b`
+ Variance :math:`\frac{1+\kappa^{4}}{b^2\kappa^2 }`
+ ======== ========================
Parameters
----------
- b:
+ b: float
Scale parameter (b > 0)
- kappa:
+ kappa: float
Symmetry parameter (kappa > 0)
+ mu: float
+ Location parameter
- See also: https://en.wikipedia.org/wiki/Asymmetric_Laplace_distribution
+ See Also:
+ --------
+ `Reference <https://en.wikipedia.org/wiki/Asymmetric_Laplace_distribution>`_
"""
- def __init__(self, b, kappa, testval=0.0, *args, **kwargs):
- self.b = tt.as_tensor_variable(b)
- self.kappa = tt.as_tensor_variable(kappa)
+ def __init__(self, b, kappa, mu=0, *args, **kwargs):
+ self.b = tt.as_tensor_variable(floatX(b))
+ self.kappa = tt.as_tensor_variable(floatX(kappa))
+ self.mu = mu = tt.as_tensor_variable(floatX(mu))
- super().__init__(*args, **kwargs, testval=testval)
+ self.mean = self.mu - (self.kappa - 1 / self.kappa) / b
+ self.variance = (1 + self.kappa ** 4) / (self.kappa ** 2 * self.b ** 2)
- def _random(self, b, kappa, size=None):
- if size is not None:
- u = np.random.uniform(size=size)
- x = -np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b) * (
- u > ((kappa ** 2) / (1 + kappa ** 2))
- ) + kappa * np.log(u * (1 + kappa ** 2) / (kappa ** 2)) / b * (
- u < ((kappa ** 2) / (1 + kappa ** 2))
- )
- return x
-
- u = np.random.uniform()
- if u > (kappa ** 2) / (1 + kappa ** 2):
- x = -np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b)
- else:
- x = kappa * np.log(u * (1 + kappa ** 2) / (kappa ** 2)) / b
+ super().__init__(*args, **kwargs)
- return x
+ def _random(self, b, kappa, size=None):
+ u = np.random.uniform(size=size)
+ switch = kappa ** 2 / (1 + kappa ** 2)
+ non_positive_x = kappa * np.log(u * (1 / switch)) / b
+ positive_x = -np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b)
+ draws = non_positive_x * (u <= switch) + positive_x * (u > switch)
+ return draws
def random(self, point=None, size=None):
"""
@@ -1741,6 +1750,7 @@ def logp(self, value):
-------
TensorVariable
"""
+ value = value - self.mu
return bound(
tt.log(self.b / (self.kappa + (self.kappa ** -1)))
+ (-value * self.b * tt.sgn(value) * (self.kappa ** tt.sgn(value))),
diff --git a/pymc3/tests/test_distributions.py b/pymc3/tests/test_distributions.py
index 948041373e..7f7e1195e9 100644
--- a/pymc3/tests/test_distributions.py
+++ b/pymc3/tests/test_distributions.py
@@ -220,10 +220,10 @@ def build_model(distfam, valuedomain, vardomains, extra_args=None):
return m
-def laplace_asymmetric_logpdf(value, kappa, b=None):
+def laplace_asymmetric_logpdf(value, kappa, b):
kapinv = 1 / kappa
- lPx = value * np.where(value >= 0, -kappa, kapinv)
- lPx -= np.log(kappa + kapinv)
+ lPx = value * b * np.where(value >= 0, -kappa, kapinv)
+ lPx += np.log(b / (kappa + kapinv))
return lPx
@@ -999,7 +999,7 @@ def test_laplace_asymmetric(self):
self.pymc3_matches_scipy(
AsymmetricLaplace,
R,
- {"b": Domain([0, 1, inf]), "kappa": Rplus},
+ {"b": Rplus, "kappa": Rplus},
laplace_asymmetric_logpdf,
)
diff --git a/pymc3/tests/test_distributions_random.py b/pymc3/tests/test_distributions_random.py
index eb359091aa..19b7901f8e 100644
--- a/pymc3/tests/test_distributions_random.py
+++ b/pymc3/tests/test_distributions_random.py
@@ -642,9 +642,7 @@ def ref_rand(size, kappa, b):
return x
- pymc3_random(
- pm.AsymmetricLaplace, {"b": Domain([0, 1, np.inf]), "kappa": Rplus}, ref_rand=ref_rand
- )
+ pymc3_random(pm.AsymmetricLaplace, {"b": Rplus, "kappa": Rplus}, ref_rand=ref_rand)
def test_lognormal(self):
def ref_rand(size, mu, tau):
From d3331418f4ac89cf57ac35beb0be688d5ae5098d Mon Sep 17 00:00:00 2001
From: Chandan Kumar Roy <chandanroy559@outlook.com>
Date: Tue, 5 Jan 2021 18:53:25 +0530
Subject: [PATCH 5/7] updated
---
pymc3/distributions/continuous.py | 19 +++++++++++--------
pymc3/tests/test_distributions.py | 5 +++--
pymc3/tests/test_distributions_random.py | 18 ++++++++----------
3 files changed, 22 insertions(+), 20 deletions(-)
diff --git a/pymc3/distributions/continuous.py b/pymc3/distributions/continuous.py
index 88cff6779f..c60898ab16 100644
--- a/pymc3/distributions/continuous.py
+++ b/pymc3/distributions/continuous.py
@@ -1669,14 +1669,14 @@ class AsymmetricLaplace(Continuous):
The pdf of this distribution is
.. math::
- {f(x|\\b,\kappa) =
- \left({\frac{\\b}{\kappa + 1/\kappa}}\right)\,e^{-(x)\\b\,s\kappa ^{s}}}
+ {f(x|\\b,\kappa,\mu) =
+ \left({\frac{\\b}{\kappa + 1/\kappa}}\right)\,e^{-(x-\mu)\\b\,s\kappa ^{s}}}
where
.. math::
- s = sgn(x)
+ s = sgn(x-\mu)
======== ========================
Support :math:`x \in \mathbb{R}`
@@ -1706,13 +1706,16 @@ def __init__(self, b, kappa, mu=0, *args, **kwargs):
self.mean = self.mu - (self.kappa - 1 / self.kappa) / b
self.variance = (1 + self.kappa ** 4) / (self.kappa ** 2 * self.b ** 2)
+ assert_negative_support(kappa, "kappa", "AsymmetricLaplace")
+ assert_negative_support(b, "b", "AsymmetricLaplace")
+
super().__init__(*args, **kwargs)
- def _random(self, b, kappa, size=None):
+ def _random(self, b, kappa, mu, size=None):
u = np.random.uniform(size=size)
switch = kappa ** 2 / (1 + kappa ** 2)
- non_positive_x = kappa * np.log(u * (1 / switch)) / b
- positive_x = -np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b)
+ non_positive_x = mu + kappa * np.log(u * (1 / switch)) / b
+ positive_x = mu - np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b)
draws = non_positive_x * (u <= switch) + positive_x * (u > switch)
return draws
@@ -1733,8 +1736,8 @@ def random(self, point=None, size=None):
-------
array
"""
- b, kappa = draw_values([self.b, self.kappa], point=point, size=size)
- return generate_samples(self._random, b, kappa, dist_shape=self.shape, size=size)
+ b, kappa, mu = draw_values([self.b, self.kappa, self.mu], point=point, size=size)
+ return generate_samples(self._random, b, kappa, mu, dist_shape=self.shape, size=size)
def logp(self, value):
"""
diff --git a/pymc3/tests/test_distributions.py b/pymc3/tests/test_distributions.py
index 7f7e1195e9..c2548423f6 100644
--- a/pymc3/tests/test_distributions.py
+++ b/pymc3/tests/test_distributions.py
@@ -220,8 +220,9 @@ def build_model(distfam, valuedomain, vardomains, extra_args=None):
return m
-def laplace_asymmetric_logpdf(value, kappa, b):
+def laplace_asymmetric_logpdf(value, kappa, b, mu):
kapinv = 1 / kappa
+ value = value - mu
lPx = value * b * np.where(value >= 0, -kappa, kapinv)
lPx += np.log(b / (kappa + kapinv))
return lPx
@@ -999,7 +1000,7 @@ def test_laplace_asymmetric(self):
self.pymc3_matches_scipy(
AsymmetricLaplace,
R,
- {"b": Rplus, "kappa": Rplus},
+ {"b": Rplus, "kappa": Rplus, "mu": R},
laplace_asymmetric_logpdf,
)
diff --git a/pymc3/tests/test_distributions_random.py b/pymc3/tests/test_distributions_random.py
index 19b7901f8e..6cc4fe8042 100644
--- a/pymc3/tests/test_distributions_random.py
+++ b/pymc3/tests/test_distributions_random.py
@@ -377,7 +377,7 @@ class TestLaplace(BaseTestCases.BaseTestCase):
class TestAsymmetricLaplace(BaseTestCases.BaseTestCase):
distribution = pm.AsymmetricLaplace
- params = {"kappa": 1.0, "b": 1.0}
+ params = {"kappa": 1.0, "b": 1.0, "mu": 0.0}
class TestLognormal(BaseTestCases.BaseTestCase):
@@ -632,17 +632,15 @@ def ref_rand(size, mu, b):
pymc3_random(pm.Laplace, {"mu": R, "b": Rplus}, ref_rand=ref_rand)
def test_laplace_asymmetric(self):
- def ref_rand(size, kappa, b):
+ def ref_rand(size, kappa, b, mu):
u = np.random.uniform(size=size)
- x = -np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b) * (
- u > ((kappa ** 2) / (1 + kappa ** 2))
- ) + kappa * np.log(u * (1 + kappa ** 2) / (kappa ** 2)) / b * (
- u < ((kappa ** 2) / (1 + kappa ** 2))
- )
-
- return x
+ switch = kappa ** 2 / (1 + kappa ** 2)
+ non_positive_x = mu + kappa * np.log(u * (1 / switch)) / b
+ positive_x = mu - np.log((1 - u) * (1 + kappa ** 2)) / (kappa * b)
+ draws = non_positive_x * (u <= switch) + positive_x * (u > switch)
+ return draws
- pymc3_random(pm.AsymmetricLaplace, {"b": Rplus, "kappa": Rplus}, ref_rand=ref_rand)
+ pymc3_random(pm.AsymmetricLaplace, {"b": Rplus, "kappa": Rplus, "mu": R}, ref_rand=ref_rand)
def test_lognormal(self):
def ref_rand(size, mu, tau):
From a19c8f9c5bf04fe4c30feb3a653c4a2b764b4141 Mon Sep 17 00:00:00 2001
From: Chandan Kumar Roy <chandanroy559@outlook.com>
Date: Tue, 5 Jan 2021 19:52:20 +0530
Subject: [PATCH 6/7] RELEASE-NOTES.md and continuous.rst updated
---
RELEASE-NOTES.md | 2 +-
docs/source/api/distributions/continuous.rst | 1 +
2 files changed, 2 insertions(+), 1 deletion(-)
diff --git a/RELEASE-NOTES.md b/RELEASE-NOTES.md
index a313d6cc33..4cbfb52e73 100644
--- a/RELEASE-NOTES.md
+++ b/RELEASE-NOTES.md
@@ -21,7 +21,7 @@ It also brings some dreadfully awaited fixes, so be sure to go through the chang
- `plot_posterior_predictive_glm` now works with `arviz.InferenceData` as well (see [#4234](https://github.com/pymc-devs/pymc3/pull/4234))
- Add `logcdf` method to all univariate discrete distributions (see [#4387](https://github.com/pymc-devs/pymc3/pull/4387)).
- Add `random` method to `MvGaussianRandomWalk` (see [#4388](https://github.com/pymc-devs/pymc3/pull/4388))
-
+- `AsymmetricLaplace` distribution added (see [#4392](https://github.com/pymc-devs/pymc3/pull/4392)).
### Maintenance
- Fixed bug whereby partial traces returns after keyboard interrupt during parallel sampling had fewer draws than would've been available [#4318](https://github.com/pymc-devs/pymc3/pull/4318)
- Make `sample_shape` same across all contexts in `draw_values` (see [#4305](https://github.com/pymc-devs/pymc3/pull/4305)).
diff --git a/docs/source/api/distributions/continuous.rst b/docs/source/api/distributions/continuous.rst
index fcc49d2e11..9222dcb797 100644
--- a/docs/source/api/distributions/continuous.rst
+++ b/docs/source/api/distributions/continuous.rst
@@ -16,6 +16,7 @@ Continuous
Kumaraswamy
Exponential
Laplace
+ AsymmetricLaplace
StudentT
HalfStudentT
Cauchy
From 39146c78edb097951d4eaddf450641b70b792906 Mon Sep 17 00:00:00 2001
From: Chandan Kumar Roy <chandanroy559@outlook.com>
Date: Tue, 5 Jan 2021 19:56:24 +0530
Subject: [PATCH 7/7]
---
RELEASE-NOTES.md | 1 +
1 file changed, 1 insertion(+)
diff --git a/RELEASE-NOTES.md b/RELEASE-NOTES.md
index 4cbfb52e73..10c3e5a2ec 100644
--- a/RELEASE-NOTES.md
+++ b/RELEASE-NOTES.md
@@ -22,6 +22,7 @@ It also brings some dreadfully awaited fixes, so be sure to go through the chang
- Add `logcdf` method to all univariate discrete distributions (see [#4387](https://github.com/pymc-devs/pymc3/pull/4387)).
- Add `random` method to `MvGaussianRandomWalk` (see [#4388](https://github.com/pymc-devs/pymc3/pull/4388))
- `AsymmetricLaplace` distribution added (see [#4392](https://github.com/pymc-devs/pymc3/pull/4392)).
+
### Maintenance
- Fixed bug whereby partial traces returns after keyboard interrupt during parallel sampling had fewer draws than would've been available [#4318](https://github.com/pymc-devs/pymc3/pull/4318)
- Make `sample_shape` same across all contexts in `draw_values` (see [#4305](https://github.com/pymc-devs/pymc3/pull/4305)).