Source code for hierarc.Util.likelihood_util

import numpy as np
from scipy.stats import truncnorm




def log_likelihood_cov(data, model, cov_error):
    """Log likelihood of the data given a model.

    :param data: data vector
    :param model: model vector
    :param cov_error: inverse covariance matrix
    :return: log likelihood
    """
    delta = data - model
    return -delta.dot(cov_error.dot(delta)) / 2.0





def cov_error_create(error_independent, error_covariance):
    """Generates an error covariance matrix from a set of independent uncertainties
    combined with a fully covariant term.

    :param error_independent: array of Gaussian 1-sigma uncertainties
    :param error_covariance: float, shared covariant error among all data points. So if
        all data points are off by 1-sigma, then the log likelihood is 1-sigma
    :return: error covariance matrix
    """
    error_covariance_array = np.ones_like(error_independent) * error_covariance
    error = np.outer(error_covariance_array, error_covariance_array) + np.diag(
        error_independent**2
    )
    return np.linalg.inv(error)





def get_truncated_normal(mean=0, sd=1, low=0, upp=10, size=1):
    """

    :param mean: mean of normal distribution
    :param sd: standard deviation
    :param low: lower bound
    :param upp: upper bound
    :return: float, draw of distribution
    """
    return truncnorm((low - mean) / sd, (upp - mean) / sd, loc=mean, scale=sd).rvs(size)