From 8be9b27b5497bd0f357c44baa834be3af9b6d646 Mon Sep 17 00:00:00 2001
From: Colm Talbot <colm.talbot@ligo.org>
Date: Mon, 17 Sep 2018 21:54:20 +1000
Subject: [PATCH] move log_likelihood methods
---
tupak/core/likelihood.py | 59 ++++++++++++++++++++--------------------
1 file changed, 30 insertions(+), 29 deletions(-)
diff --git a/tupak/core/likelihood.py b/tupak/core/likelihood.py
index 1ac787460..460ae7c55 100644
--- a/tupak/core/likelihood.py
+++ b/tupak/core/likelihood.py
@@ -153,15 +153,15 @@ class GaussianLikelihood(Analytical1DLikelihood):
if self.sigma is None:
self.parameters['sigma'] = None
- def __repr__(self):
- return self.__class__.__name__ + '(x={}, y={}, func={}, sigma={})'\
- .format(self.x, self.y, self.func.__name__, self.sigma)
-
def log_likelihood(self):
log_l = np.sum(- (self.residual / self.sigma)**2 / 2
- np.log(2 * np.pi * self.sigma) / 2)
return log_l
+ def __repr__(self):
+ return self.__class__.__name__ + '(x={}, y={}, func={}, sigma={})' \
+ .format(self.x, self.y, self.func.__name__, self.sigma)
+
@property
def sigma(self):
"""
@@ -205,6 +205,20 @@ class PoissonLikelihood(Analytical1DLikelihood):
Analytical1DLikelihood.__init__(self, x=x, y=y, func=func)
+ def log_likelihood(self):
+ rate = self.func(self.x, **self.model_parameters)
+ if not isinstance(rate, np.ndarray):
+ raise ValueError(
+ "Poisson rate function returns wrong value type! "
+ "Is {} when it should be numpy.ndarray".format(type(rate)))
+ elif np.any(rate < 0.):
+ raise ValueError(("Poisson rate function returns a negative",
+ " value!"))
+ elif np.any(rate == 0.):
+ return -np.inf
+ else:
+ return np.sum(-rate + self.y * np.log(rate) - gammaln(self.y + 1))
+
def __repr__(self):
return Analytical1DLikelihood.__repr__(self)
@@ -222,19 +236,6 @@ class PoissonLikelihood(Analytical1DLikelihood):
raise ValueError("Data must be non-negative integers")
self.__y = y
- def log_likelihood(self):
- rate = self.func(self.x, **self.model_parameters)
- if not isinstance(rate, np.ndarray):
- raise ValueError("Poisson rate function returns wrong value type! "
- "Is {} when it should be numpy.ndarray".format(type(rate)))
- elif np.any(rate < 0.):
- raise ValueError(("Poisson rate function returns a negative",
- " value!"))
- elif np.any(rate == 0.):
- return -np.inf
- else:
- return np.sum(-rate + self.y * np.log(rate)) - np.sum(gammaln(self.y + 1))
-
class ExponentialLikelihood(Analytical1DLikelihood):
def __init__(self, x, y, func):
@@ -255,6 +256,12 @@ class ExponentialLikelihood(Analytical1DLikelihood):
"""
Analytical1DLikelihood.__init__(self, x=x, y=y, func=func)
+ def log_likelihood(self):
+ mu = self.func(self.x, **self.model_parameters)
+ if np.any(mu < 0.):
+ return -np.inf
+ return -np.sum(np.log(mu) + (self.y / mu))
+
def __repr__(self):
return Analytical1DLikelihood.__repr__(self)
@@ -271,12 +278,6 @@ class ExponentialLikelihood(Analytical1DLikelihood):
raise ValueError("Data must be non-negative")
self.__y = y
- def log_likelihood(self):
- mu = self.func(self.x, **self.model_parameters)
- if np.any(mu < 0.):
- return -np.inf
- return -np.sum(np.log(mu) + (self.y / mu))
-
class StudentTLikelihood(Analytical1DLikelihood):
def __init__(self, x, y, func, nu=None, sigma=1.):
@@ -317,6 +318,12 @@ class StudentTLikelihood(Analytical1DLikelihood):
if self.nu is None:
self.parameters['nu'] = None
+ def log_likelihood(self):
+ if self.__get_nu() <= 0.:
+ raise ValueError("Number of degrees of freedom for Student's t-likelihood must be positive")
+
+ return self.__summed_log_likelihood(self.__get_nu())
+
def __repr__(self):
return self.__class__.__name__ + '(x={}, y={}, func={}, nu={}, sigma={})'\
.format(self.x, self.y, self.func.__name__, self.nu, self.sigma)
@@ -326,12 +333,6 @@ class StudentTLikelihood(Analytical1DLikelihood):
""" Converts 'scale' to 'precision' """
return 1. / self.sigma ** 2
- def log_likelihood(self):
- if self.__get_nu() <= 0.:
- raise ValueError("Number of degrees of freedom for Student's t-likelihood must be positive")
-
- return self.__summed_log_likelihood(self.__get_nu())
-
def __get_nu(self):
""" This checks if nu or sigma have been set in parameters. If so, those
values will be used. Otherwise, the attribute nu is used. The logic is
--
GitLab