API Reference

Graph

The graph interface carrying the main UI.

Graph

Graph Construction

The state indexing utility for mapping between between the integer vectors of states and the model properties they represent.

StateIndexer	Manages multiple independent property sets and slots with attribute access.
PropertySet	Collection of properties forming a combinatorial state space.
Property	Defines a single property in a state space.
PropertyDict	Dict subclass that provides attribute access to keys.
StateVector	Dict-like interface for multi-PropertySet state with attribute access.

Inference

The main inference UI is Graph.svgd, which internally uses the SVGD class.

SVGD	Stein Variational Gradient Descent (SVGD) for Bayesian parameter inference.

Sparse observations

Sparse observations for multi-feature data.

SparseObservations	Sparse representation of multivariate observations.
dense_to_sparse	Convert dense NaN-padded array to sparse observation format.
is_sparse_observations	Check if data is in sparse observation format.

Priors

Prior distributions of model parameters.

DataPrior	Data-informed prior estimated from observed data.
Prior	Base class for prior distributions.
GaussPrior	Gaussian prior distribution.
HalfCauchyPrior	Half-Cauchy prior distribution (positive support only).

Learning rate and regularization

Learning rate schedules and regularization techniques for Stein Variational Gradient Descent.

StepSizeSchedule	Base class for step size schedules.
ConstantStepSize	Constant step size (default behavior).
ExpStepSize	Exponential decay schedule: step_size = first_step * exp(-iteration/tau) + last_step * (1 - exp(-iteration/tau)).
AdaptiveStepSize	Adaptive step size based on particle spread (KL divergence proxy).
WarmupExpStepSize	Linear warmup followed by exponential decay.

Regularization

Moment regularization techniques for Stein Variational Gradient Descent.

RegularizationSchedule	Base class for regularization schedules.
ConstantRegularization	Constant regularization (default behavior).
ExpRegularization	Exponential decay schedule: reg = first_reg * exp(-iteration/tau) + last_reg * (1 - exp(-iteration/tau)).

Preconditioning

Automated preconditioning for homogenous learning.

MomentJacobianPreconditioner	Moment Jacobian preconditioner for multi-scale SVGD.
FisherPreconditioner	Diagonal Fisher information preconditioner for multi-scale SVGD.

Optimizers

Adaptive optimizers. Recommended only for models that will not otherwise converge.

Adam	Adam optimizer for SVGD with per-parameter adaptive learning rates.
SGDMomentum	SGD with momentum optimizer for SVGD.
RMSprop	RMSprop optimizer for SVGD.
Adagrad	Adagrad optimizer for SVGD.

Optax optimizer support

Support for using optimizers from the external Optax library some of them pre-wrapped for convenience. Recommended only for models that will not otherwise converge, and for users who want to experiment with a wider range of optimizers.

OptaxOptimizer	Wrapper to use Optax optimizers with phasic’s SVGD interface.
optax_adam	Create Optax Adam optimizer wrapped for phasic.
optax_adamw	Create Optax AdamW optimizer wrapped for phasic.
optax_sgd	Create Optax SGD optimizer wrapped for phasic.
optax_rmsprop	Create Optax RMSprop optimizer wrapped for phasic.
optax_adagrad	Create Optax Adagrad optimizer wrapped for phasic.
optax_chain	Create chained Optax transforms wrapped for phasic.
optax_lion	Create Optax Lion optimizer wrapped for phasic.

Exceptions

Custom exceptions for error handling in phasic.

PTDAlgorithmsError	Base exception for all phasic errors.
PTDConfigError	Configuration error with suggested fixes.
PTDBackendError	Backend not available error.
PTDFeatureError	Feature not available on this platform.
PTDJAXError	JAX-specific error.

Library utilities

Configuration and Logging.

clear_caches	Clear all caching.
configure	Configure phasic globally.
get_config	Get current global configuration.
get_available_options	Get dictionary of available options on this system.
PTDAlgorithmsConfig	Global configuration for phasic behavior.
reset_config	Reset configuration to default.
set_log_level	Context manager and callable for changing the phasic logging level at runtime.
get_logger	Get a logger for the specified module.