A detailed description of mhn’s subpackages, submodules and functions#

This part of the documentation contains descriptions of all functions included in the mhn package.

Directly callable functions and constants#

mhn.set_seed(seed: int)#

Set the random seed for reproducibility. Internally, this sets the seed for the numpy random generator.

Parameters:: seed (int) – The seed value to use for random number generators.

mhn.cuda_available() → Literal[CUDA_AVAILABLE, CUDA_NOT_AVAILABLE, CUDA_NOT_FUNCTIONAL]#: Call this function if you want to know if the mhn package is able to use CUDA functions on your device.

mhn.CUDA_AVAILABLE#: Constant returned by cuda_available() if CUDA functions are available and working correctly.

mhn.CUDA_NOT_AVAILABLE#: Constant returned by cuda_available() if the CUDA compiler (nvcc) was not present during installation. If this was not expected, ensure that the CUDA toolkit is properly installed and accessible.

mhn.CUDA_NOT_FUNCTIONAL#: Constant returned by cuda_available() if the CUDA compiler (nvcc) is available, but CUDA functions are not working as expected. In this case, check your CUDA drivers and installation for potential issues.

This part of the mhn package contains functions to compute and work with MHNs on the full state-space.

The structure of this part of the package is very similar to the original R implementation by Schill et al. (2019):

Likelihood:: contains functions to compute the log-likelihood score and its gradient without state-space restriction as well as functions for matrix-vector multiplications with the transition rate matrix and [I-Q]^(-1)
ModelConstruction:: contains functions to generate random MHNs, build their transition rate matrix Q, the diagonal of Q, and to generate and independence model for a given distribution
RegularizedOptimization:: contains functions to learn an cMHN for a given data distribution, implements the L1 regularization
UtilityFunctions:: contains functions useful for preprocessing data to be used as training data
PerformanceCriticalCode:: contains functions that dominate the runtime of score and gradient computations and must therefore be implemented efficiently
fisher:: contains functions to compute the Fisher information matrix for MHN

This part of the package contains functions that are needed for model training.

This submodule contains Optimizer classes to learn an MHN from mutation data.

This submodule contains classes to represent Mutual Hazard Networks.

This module contains some utility functions for working with MHNs.

This part of the mhn package contains functions for the posterior sampling of MHNs with Markov Chain Monte Carlo.

kernels:: contains the Random-Walk Metropolis, Metropolis-Adjusted Langevin Algorithm (MALA) and simplified manifold MALA kernels and a base kernel base class. Those are responsible for the actual steps of MCMC.
mcmc:: contains the MCMC class that uses the kernels to create MCMC chains.