Negative Multinomial aggiornata

gemact/config.py

@@ -58,5 +58,7 @@
     'pareto2': 'distributions.Pareto2',
     'pareto1': 'distributions.Pareto1',
     'uniform': 'distributions.Uniform',
-    'multinomial': 'distributions.Multinomial'
+    'multinomial': 'distributions.Multinomial',
+    'dirichlet multinomial' : 'distributions.Dirichlet_Multinomial',
+    'negative multinomial' : 'distributions.NegMultinom'
 }

gemact/distributions.py

@@ -6261,7 +6261,7 @@ def name():
 
     @staticmethod
     def category():
-        return {'frequency'}
+        return {''}
 
 
     def cov(self):
@@ -6326,4 +6326,345 @@ def logpmf(self, x):
         return stats.multinomial.logpmf(n=self.n, p=self.p, x=x)
 
 
+
+class Dirichlet_Multinomial(_MultDiscreteDistribution):
+    """
+    Dirichlet Multinomial distribution.
+    Wrapper to scipy dirichlet multinomial distribution (``scipy.stats.dirichlet_multinomial``).
+    Refer to :py:class:'~__MultDiscreteDistribution' for additional details.
+
+
+    :param seed: Used to set a specific seed (default=np.random.RandomState).
+    :type seed: int
+    :param \\**kwargs:
+        See below
+
+    :Keyword Arguments:
+        * *alpha* ( ``int`` or ``numpy.ndarray``) --
+          Concentration parameters.           
+        * *n* (``int``) --
+          Number of trials.
+    """
+
+    def __init__(self, seed=None, **kwargs):
+        _DiscreteDistribution.__init__(self)
+        self.n = kwargs['n']
+        self.alpha = kwargs['alpha']
+        self.seed = seed
+
+    @property
+    def seed(self):
+        return self.__seed
+
+    @seed.setter
+    def seed(self, value):
+        if value is None:
+            value = np.random.randint(1, 1001)
+
+        hf.assert_type_value(value, 'seed', logger, (float,int))
+        value = int(value)
+        self.__seed = value
+
+    @property
+    def n(self):
+        return self.__n
+
+    @n.setter
+    def n(self, value):
+        hf.assert_type_value(value, 'n', logger, (float, int), lower_bound=1, lower_close=True)
+        value = int(value)
+        self.__n = value
+
+    @property
+    def alpha(self):
+        return self.__alpha
+
+    @alpha.setter
+    def alpha(self, value):
+        for element in value:
+            hf.assert_type_value(element, 'alpha', logger, (float, int))                
+        value = np.array(value)
+        self.__alpha = value
+
+    @property
+    def _dist(self):
+        return stats.dirichlet_multinomial(n=self.n, alpha=self.alpha, seed=self.seed)
+
+    @staticmethod
+    def name():
+        return 'dirichlet multinomial'
+
+    @staticmethod
+    def category():
+        return {''}
+
+    def cov(self):
+       """
+        Covariance Matrix of a Dirichlet Multinomial Distribution.
+
+        :return: Covariance Matrix.
+        :rtype: ``float``
+        """
+
+       return stats.dirichlet_multinomial.cov(n=self.n, alpha=self.alpha)
+
+
+    def var(self):
+       """
+        Variances of a Dirichlet Multinomial Distribution.
+
+        :return: Array of Variances.
+        :rtype: numpy.ndarray
+        """
+
+       return np.diag(stats.dirichlet_multinomial.cov(n=self.n, alpha=self.alpha))
+
+    def pmf(self, x):
+        """
+        Probability mass function of the Dirichlet Multinomial Distribution.
+
+        :param x: quantile where probability mass function is evaluated.
+        :type x: ``int``
+
+        :return: probability mass function.
+        :rtype: ``numpy.float64`` or ``numpy.ndarray``
+        """
+
+        if sum(x) != self.n:
+         raise ValueError("n != sum(x), i.e. one is wrong")    
+        return stats.dirichlet_multinomial.pmf(n=self.n, alpha=self.alpha, x=x)
+
+    def logpmf(self, x):
+        """
+        Natural logarithm of the probability mass function of the Dirichlet Multinomial Distribution.
+
+        :param x: quantile where the (natural) probability mass function logarithm is evaluated.
+        :type x: ``int``
+        :return: natural logarithm of the probability mass function
+        :rtype: ``numpy.float64`` or ``numpy.ndarray``
+        """
+
+        if sum(x) != self.n:
+         raise ValueError("n != sum(x), i.e. one is wrong")       
+        return stats.dirichlet_multinomial.logpmf(n=self.n, alpha=self.alpha, x=x)
+
+
+    def mean(self):
+        """
+        Mean of the Dirichlet Multinomial Distribution.
+
+        :param x: quantile where the (natural) probability mass function logarithm is evaluated.
+        :type x: ``int``
+        :return: natural logarithm of the probability mass function
+        :rtype: ``numpy.float64`` or ``numpy.ndarray``
+        """        
+        return stats.dirichlet_multinomial.mean(n=self.n, alpha=self.alpha)
+
+
+    def rvs(self, size=1, random_state=None):
+        """
+        Random variates generator function.
+
+        :param size: random variates sample size (default is 1).
+        :type size: ``int``, optional
+        :param random_state: random state for the random number generator.
+        :type random_state: ``int``, optional
+        :return: random variates.
+        :rtype: ``numpy.int`` or ``numpy.ndarray``
+
+        """
+
+        random_state = hf.handle_random_state(random_state, logger)
+        np.random.seed(random_state)
+        hf.assert_type_value(size, 'size', logger, (float, int), lower_bound=1)
+        size = int(size)       
+        alpha = self.alpha
+        n = self.n    
+        if isinstance(alpha, np.ndarray) and len(alpha.shape) == 1:
+            alpha = np.tile(alpha, (size, 1))
+            n = np.full(size, n)                
+        G = np.random.gamma(shape=alpha, scale=1.0)
+        prob = G / np.sum(G, axis=1, keepdims=True)        
+        ridx = np.sum(G, axis=1) == 0        
+        if np.any(ridx):
+            for i in np.where(ridx)[0]:
+                prob[i, :] = np.random.multinomial(1, alpha[i, :] / np.sum(alpha[i, :]), n=1).flatten()                
+        rdm = np.array([np.random.multinomial(n[i], prob[i, :]) for i in range(size)])      
+        return rdm
+
+
+class NegMultinom(_MultDiscreteDistribution):
+
+    """
+    Negative Multinomial distribution.
+
+    :param loc: Location parameter to shift the support (default=0).
+    :type loc: ``int``, optional
+
+    :param \\**kwargs:
+        See below
+
+    :Keyword Arguments:
+        * *x0* (``int``) --
+          Size parameter of the negative multinomial distribution.
+        * *p* (``float``) --
+          Probability parameter of the negative multinomial distribution.
+
+    """
+
+    def __init__(self, x0, p, loc=0):
+        _DiscreteDistribution.__init__(self)
+        self.x0 = x0
+        self.p = p
+        self.loc = loc
+
+    @property
+    def x0(self):
+        return self.__x0
+
+    @x0.setter
+    def x0(self, value):
+        hf.assert_type_value(value, 'x0', logger, (float, int), lower_bound=0, lower_close=False)
+        self.__x0 = value
+
+    @property
+    def p(self):
+        return self.__p
+
+    @p.setter
+    def p(self, value):
+        value = np.array(value)
+        for element in value:
+            hf.assert_type_value(
+                element, 'p', logger, (float, np.floating), 
+                lower_bound=0, upper_bound=1, lower_close=True, upper_close=True
+            )
+
+        if np.sum(value) >= 1:
+            raise ValueError("Sum of success probabilities must be less than 1")
+
+        self.__p = value
+        self.__p0 = 1 - np.sum(value)  # Failure probability
+
+    @property
+    def p0(self):
+        """Probability of failure (computed as 1 - sum(p))"""
+        return self.__p0
+
+    @staticmethod
+    def name():
+        return 'negative multinomial'
+
+    @staticmethod
+    def category():
+        return {''}
+
+    def pmf(self, x):
+        """
+        Probability mass function.
+
+        PMF formula from reference:
+        Γ(∑x_i) * p0^x0 / Γ(x0) * ∏(p_i^x_i / x_i!)
+
+        :param x: Quantile where PMF is evaluated.
+        :type x: ``numpy.ndarray``
+        :return: Probability mass function evaluated at x.
+        :rtype: ``numpy.float64``
+        """
+        x = np.array(x)
+        x0 = self.x0
+
+        gamma_term = (special.gamma(np.sum(x)+x0) / special.gamma(x0))
+        prob_term = (self.p0 ** x0) * np.prod((self.p ** x) / special.factorial(x))
+
+        return gamma_term * prob_term
+
+
+    def logpmf(self, x):
+        """
+        Natural logarithm of the probability mass function.
+
+        :param x: Quantile where log-PMF is evaluated.
+        :type x: ``numpy.ndarray``
+        :return: Log of probability mass function evaluated at x.
+        :rtype: ``numpy.float64``
+        """
+        return np.log(self.pmf(x))
+
+    def mean(self):
+        """
+        Mean vector of the distribution.
+
+        :return: Mean vector.
+        :rtype: ``numpy.ndarray``
+        """
+        return (self.x0 / self.p0) * self.p
+
+    def var(self):
+        """
+        Variances of a Negative Multinomial Distribution.
+
+        :return: Array of Variances.
+        :rtype: ``numpy.ndarray``
+        """
+        return (self.x0 / self.p0**2) * self.p**2 + (self.x0 / self.p0) * self.p
+
+    def cov(self):
+        """
+        Covariance matrix of a Negative Multinomial Distribution.
+
+        :return: Covariance matrix.
+        :rtype: ``numpy.ndarray``
+        """
+        p = self.p
+        x0 = self.x0
+        p0 = self.p0
+
+        # Diagonal terms
+        diag = (x0 / p0**2) * p**2 + (x0 / p0) * p
+
+        # Off-diagonal terms
+        off_diag = (x0 / p0**2) * np.outer(p, p)
+
+        # Create full covariance matrix
+        cov_matrix = np.diag(diag) + off_diag - np.diag(np.diag(off_diag))
+
+        return cov_matrix
+
+    def rvs(self, size=1, random_state=None):
+        """
+        Random variates generator function.
+
+        :param size: Number of random variates to generate (default=1).
+        :type size: ``int``
+        :param random_state: Random state for reproducibility.
+        :type random_state: ``int``, optional
+        :return: Random variates.
+        :rtype: ``numpy.ndarray``
+        """
+        random_state = hf.handle_random_state(random_state, logger)
+        np.random.seed(random_state)
+
+        samples = []
+        for _ in range(size):
+            total = np.random.negative_binomial(self.x0, self.p0)
+            if total > 0:
+                counts = np.random.multinomial(total, self.p / (1 - self.p0))
+            else:
+                counts = np.zeros_like(self.p)
+            samples.append(counts)
+
+        return np.array(samples)
+
+    def mgf(self, t):
+        """
+        Moment generating function.
+
+        :param t: Vector where MGF is evaluated.
+        :type t: ``numpy.ndarray``
+        :return: Moment generating function evaluated at t.
+        :rtype: ``numpy.float64``
+        """
+        exponent = np.sum(self.p * np.exp(t))
+        return (self.p0 / (1 - exponent)) ** self.x0
 

gemact/libraries.py

@@ -13,3 +13,5 @@
 import timeit
 from twiggy import quick_setup, log
 from itertools import groupby
+from scipy.special import gammaln
+import math as mt