High-level description
This file contains unit tests for theIIDExponentialBayesian class, which is a branch length estimator in the Cassiopeia library. The tests primarily focus on comparing the estimator’s output with numerically computed values for log likelihood, log joints, and posterior time distributions.
Code Structure
The code defines several helper functions for numerical calculations and error handling, followed by aTestIIDExponentialBayesian class containing multiple test methods. Each test method focuses on a specific aspect of the IIDExponentialBayesian class, such as comparing its output to closed-form solutions or numerical approximations.
References
This code references theIIDExponentialBayesian class from cassiopeia.tools and the CassiopeiaTree class from cassiopeia.data.
Symbols
TestIIDExponentialBayesian
Description
A class containing unit tests for theIIDExponentialBayesian branch length estimator.
Inputs
This class does not take any inputs.Outputs
This class does not return any outputs.Internal Logic
The class contains several test methods, each designed to validate different aspects of theIIDExponentialBayesian class:
-
test_against_closed_form_solution_small: This test compares the estimator’s output to closed-form solutions for a small tree with one internal node. It tests various hyperparameter settings and validates the log likelihood, log joints, posterior time distribution, and posterior mean. -
test_against_closed_form_solution_medium: Similar to the previous test, but uses a slightly larger tree with two internal nodes. This test is marked as slow and only runs with one set of hyperparameters. -
test_small_discretization_level_raises_error: This test ensures that the estimator raises aValueErrorwhen the discretization level is too small. -
test_invalid_tree_topology_raises_error: This test checks that the estimator raises aValueErrorfor invalid tree topologies, such as trees that are not rooted or have internal nodes with degrees other than 3. -
test_invalid_sampling_probability_raises_error: This test ensures that the estimator raises aValueErrorfor invalid sampling probabilities (outside the range (0, 1]).
Side Effects
This class does not have any side effects.calc_exact_log_full_joint
Description
Calculates the exact log full joint probability density of the observed tree topology, state vectors, and branch lengths.Inputs
Outputs
Internal Logic
The function iterates through the edges of the tree and calculates the likelihood contribution from both the birth process and the mutation process. It considers the sampling probability and uses binomial coefficients to account for the number of mutations along each edge.calc_numerical_log_likelihood
Description
Calculates the numerical log likelihood of the observed data (character states and tree topology) using numerical integration.Inputs
Outputs
Internal Logic
The function defines an inner functionf that calculates the full joint probability for a given set of internal node times. It then uses scipy.integrate.nquad to numerically integrate f over all possible internal node times, effectively marginalizing out the times and obtaining the likelihood of the observed data.
calc_numerical_log_joints
Description
Calculates the numerical log joint probability density of the observed data and all possible times for a specific node in the tree.Inputs
Outputs
Internal Logic
The function iterates through all possible discretized times for the specified node. For each time, it defines an inner functionf that calculates the full joint probability for a given set of times for the remaining internal nodes. It then uses scipy.integrate.nquad to numerically integrate f over all possible times for the other nodes, effectively marginalizing out their times and obtaining the joint probability of the observed data and the specified node’s time.
numerical_posterior_time
Description
Calculates the numerical posterior time distribution for a specific node in the tree.Inputs
Outputs
Internal Logic
The function first calculates the numerical log joints usingcalc_numerical_log_joints. It then normalizes the log joints to obtain the posterior time distribution.