CUQIpy 1.0.0.post0.dev371__tar.gz → 1.0.0.post0.dev387__tar.gz

{cuqipy-1.0.0.post0.dev371 → cuqipy-1.0.0.post0.dev387}/CUQIpy.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: CUQIpy
-Version: 1.0.0.post0.dev371
+Version: 1.0.0.post0.dev387
 Summary: Computational Uncertainty Quantification for Inverse problems in Python
 Maintainer-email: "Nicolai A. B. Riis" <nabr@dtu.dk>, "Jakob S. Jørgensen" <jakj@dtu.dk>, "Amal M. Alghamdi" <amaal@dtu.dk>, Chao Zhang <chaz@dtu.dk>
 License:                                  Apache License

{cuqipy-1.0.0.post0.dev371 → cuqipy-1.0.0.post0.dev387}/CUQIpy.egg-info/SOURCES.txt

@@ -41,6 +41,7 @@ cuqi/distribution/_lognormal.py
 cuqi/distribution/_modifiedhalfnormal.py
 cuqi/distribution/_normal.py
 cuqi/distribution/_posterior.py
+cuqi/distribution/_smoothed_laplace.py
 cuqi/distribution/_uniform.py
 cuqi/experimental/__init__.py
 cuqi/experimental/mcmc/__init__.py

{cuqipy-1.0.0.post0.dev371 → cuqipy-1.0.0.post0.dev387}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: CUQIpy
-Version: 1.0.0.post0.dev371
+Version: 1.0.0.post0.dev387
 Summary: Computational Uncertainty Quantification for Inverse problems in Python
 Maintainer-email: "Nicolai A. B. Riis" <nabr@dtu.dk>, "Jakob S. Jørgensen" <jakj@dtu.dk>, "Amal M. Alghamdi" <amaal@dtu.dk>, Chao Zhang <chaz@dtu.dk>
 License:                                  Apache License

{cuqipy-1.0.0.post0.dev371 → cuqipy-1.0.0.post0.dev387}/cuqi/_version.py

@@ -8,11 +8,11 @@ import json
 
 version_json = '''
 {
- "date": "2024-07-05T10:04:17+0300",
+ "date": "2024-07-05T10:30:02+0300",
  "dirty": false,
  "error": null,
- "full-revisionid": "62f642d707e561a7c9deb0006625b11866043b81",
- "version": "1.0.0.post0.dev371"
+ "full-revisionid": "192b464555ba1f70ce8b0041ecd49ff4978d5689",
+ "version": "1.0.0.post0.dev387"
 }
 '''  # END VERSION_JSON
 

{cuqipy-1.0.0.post0.dev371 → cuqipy-1.0.0.post0.dev387}/cuqi/distribution/__init__.py

@@ -9,6 +9,7 @@ from ._gmrf import GMRF
 from ._inverse_gamma import InverseGamma
 from ._lmrf import LMRF
 from ._laplace import Laplace
+from ._smoothed_laplace import SmoothedLaplace
 from ._lognormal import Lognormal
 from ._normal import Normal
 from ._posterior import Posterior

{cuqipy-1.0.0.post0.dev371 → cuqipy-1.0.0.post0.dev387}/cuqi/distribution/_custom.py

@@ -50,13 +50,13 @@ class UserDefinedDistribution(Distribution):
         if self.logpdf_func is not None:
             return self.logpdf_func(x)
         else:
-            raise Exception("logpdf_func is not defined.")
+            raise NotImplementedError("logpdf_func is not defined.")
     
     def _gradient(self, x):
         if self.gradient_func is not None:
             return self.gradient_func(x)
         else:
-            raise Exception("gradient_func is not defined.")
+            raise NotImplementedError("gradient_func is not defined.")
 
     def _sample(self, N=1, rng=None):
         #TODO(nabr) allow sampling more than 1 sample and potentially rng?

cuqipy-1.0.0.post0.dev387/cuqi/distribution/_smoothed_laplace.py

@@ -0,0 +1,95 @@
+import numpy as np
+from cuqi.utilities import force_ndarray
+from cuqi.distribution import Distribution
+
+class SmoothedLaplace(Distribution):
+    """ Smoothed Laplace distribution. 
+
+    Defines a smoothed Laplace distribution given a location, a scale and a smoothing parameter
+    beta. The smoothed Laplace distribution is defined as
+
+    .. math::
+
+        p(x) = \\frac{1}{2b} \exp\left(-\\frac{\sqrt{(x-\mu)^2 + \beta}}{b}\\right),
+
+    where :math:`\mu` is the location (mean), :math:`b` is the scale (decay) parameter and
+    :math:`\beta` is the smoothing parameter.
+
+    The rate parameter is defined as :math:`\lambda = \\frac{1}{b}`.
+
+    The variables of this Laplace distribution are independent identically distributed (i.i.d.).
+
+    Parameters
+    ----------
+    location : scalar, list, tuple, or ndarray
+        The location parameter of the distribution.
+
+    scale : scalar, list, tuple, or ndarray
+        The scale parameter of the distribution.
+    
+    beta : scalar
+        The smoothing parameter of the distribution.
+
+    """
+
+    def __init__(self, location=None, scale=None, beta=1e-3, **kwargs):
+        super().__init__(**kwargs)
+
+        self.location = location
+        self.scale = scale
+        self.beta = beta
+
+    @property
+    def location(self):
+        """ Location parameter """
+        return self._location
+    
+    @location.setter
+    def location(self, value):
+        self._location = force_ndarray(value, flatten=True)
+
+    @property
+    def scale(self):
+        """ Scale parameter """
+        return self._scale
+    
+    @scale.setter
+    def scale(self, value):
+        self._scale = force_ndarray(value, flatten=True)
+
+    @property
+    def beta(self):
+        """ Beta parameter """
+        return self._beta
+    
+    @beta.setter
+    def beta(self, value):
+        self._beta = value
+
+    def logpdf(self, x):
+        """
+        Computes the logarithm of the probability density function at the given values of x.
+        """
+        # x accepts scalar, list, tuple, or ndarray
+        if isinstance(x, (float, int)):
+            x = np.array([x])
+        elif isinstance(x, (list, tuple)):
+            x = np.array(x)
+        return np.sum(np.log(0.5 / self.scale)) - np.sum(np.sqrt((x - self.location) ** 2 + self.beta) / self.scale)
+
+    def gradient(self, x):
+        """
+        Computes the gradient of logpdf at the given values of x.
+        """
+        # x accepts scalar, list, tuple, or ndarray
+        if isinstance(x, (float, int)):
+            x = np.array([x])
+        elif isinstance(x, (list, tuple)):
+            x = np.array(x)
+        return -np.array((x - self.location) / self.scale / np.sqrt((x - self.location) ** 2 + self.beta))
+
+    def _sample(self, N=1, rng=None):
+        """
+        Generates random samples from the distribution.
+        """
+        raise NotImplementedError(f"sample is not implemented for {self.__class__.__name__}.")

{cuqipy-1.0.0.post0.dev371 → cuqipy-1.0.0.post0.dev387}/cuqi/problem/_problem.py

@@ -710,8 +710,7 @@ class BayesianProblem(object):
         # Require gradient?
         if must_have_gradient:
             try: 
-                posterior.prior.gradient(np.zeros(posterior.prior.dim))
-                posterior.likelihood.gradient(np.zeros(posterior.likelihood.dim))
+                posterior.posterior.gradient(np.zeros(posterior.posterior.dim))
                 G = True
             except (NotImplementedError, AttributeError):
                 G = False

{cuqipy-1.0.0.post0.dev371 → cuqipy-1.0.0.post0.dev387}/tests/test_distribution.py

@@ -779,4 +779,30 @@ def test_MHN_regression(alpha, beta, gamma, expected_logpdf, expected_gradient):
     logpdf = dist.logpdf(np.array([1.0, 2.0, 3.0, 4.0, 5.0]))
     gradient = dist._gradient(np.array([1.0, 2.0, 3.0, 4.0, 5.0]))
     assert np.allclose( logpdf, np.array(expected_logpdf))
-    assert np.allclose( gradient, np.array(expected_gradient))
+    assert np.allclose( gradient, np.array(expected_gradient))
+
+def test_Smoothed_Laplace():
+    """ Test Smoothed Laplace distribution logpdf and gradient """
+
+    location = np.array([1, 2])
+    scale = np.array([1, 2])
+    scalar_laplace_0 = cuqi.distribution.Laplace(location[0], scale[0])
+
+    scalar_smoothed_laplace_0 = cuqi.distribution.SmoothedLaplace(location[0], scale[0], 1e-8)
+    scalar_smoothed_laplace_1 = cuqi.distribution.SmoothedLaplace(location[1], scale[1], 1e-8)
+    vector_smoothed_laplace = cuqi.distribution.SmoothedLaplace(location, scale, 1e-8)
+
+    x = np.array([3, 4])
+
+    # logpdf (scalar Laplace vs scalar Smoothed Laplace)
+    assert np.allclose(scalar_laplace_0.logpdf([x[0]]), scalar_smoothed_laplace_0.logpdf(x[0]))
+
+    # logpdf (scalar Smoothed Laplace * scalar Smoothed Laplace vs vector Smoothed Laplace)
+    assert np.allclose(scalar_smoothed_laplace_0.logpdf(x[0])+scalar_smoothed_laplace_1.logpdf(x[1]),
+                       vector_smoothed_laplace.logpdf(x))
+    
+    # gradient (scalar Smoothed Laplace vs analytical)
+    assert np.allclose(scalar_smoothed_laplace_0.gradient(x[0]), -1/scale[0])
+
+    # gradient (vector Smoothed Laplace vs analytical)
+    assert np.allclose(vector_smoothed_laplace.gradient(x), -1/scale)