Settings#
Settings for the PhaseGen application.
- class Settings[source]#
Bases:
object-
flatten_block_counting:
bool= True# Whether to flatten the block-counting state space when possible. In certain cases, this can be achieved by computing block probabilities and adjusting the rewards of the lineage-counting state space accordingly. This can substantially speed up computations.
-
use_pbar:
bool= False# Whether to show a progress bar for certain operations such as state space generation.
-
parallelize:
bool= False# Whether to parallelize phase-type computations across multiple CPU cores. This may improve performance in some cases, but can also be detrimental due to inter-process data copying and can lead to hanging processes.
-
regularize:
bool= True# Whether to regularize the intensity matrix for numerical stability.
-
cache_epochs:
bool= True# Whether to cache the rate matrix for different epochs which increases performance.
-
use_numba:
bool= True# Whether to use the numba-accelerated state-space construction when numba is available. Set to
Falseto force the pure-Python construction path.
-
expm_action_min_dim:
int= 1500# Van Loan matrix dimension (
(k + 1) * n_states) at or above which moments are computed via the sparse matrix-exponential action (Krylov/Taylor) instead of forming the dense propagator. The action exploits the sparsity of the rate matrix and is much faster for large state spaces, but slower for small ones. Set to a very large value to always use the dense path, or to 0 to always use the action.
-
flatten_block_counting: