code
stringlengths
66
870k
docstring
stringlengths
19
26.7k
func_name
stringlengths
1
138
language
stringclasses
1 value
repo
stringlengths
7
68
path
stringlengths
5
324
url
stringlengths
46
389
license
stringclasses
7 values
def dual_avg_step(fix_L, update_da): """does one step of the dynamics and updates the estimate of the posterior size and optimal stepsize""" def step(iteration_state, weight_and_key): mask, rng_key = weight_and_key ( previous_state, params, ...
does one step of the dynamics and updates the estimate of the posterior size and optimal stepsize
dual_avg_step
python
blackjax-devs/blackjax
blackjax/adaptation/adjusted_mclmc_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/adjusted_mclmc_adaptation.py
Apache-2.0
def adjusted_mclmc_make_adaptation_L( kernel, frac, Lfactor, max="avg", eigenvector=None ): """determine L by the autocorrelations (around 10 effective samples are needed for this to be accurate)""" def adaptation_L(state, params, num_steps, key): num_steps = int(num_steps * frac) adaptatio...
determine L by the autocorrelations (around 10 effective samples are needed for this to be accurate)
adjusted_mclmc_make_adaptation_L
python
blackjax-devs/blackjax
blackjax/adaptation/adjusted_mclmc_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/adjusted_mclmc_adaptation.py
Apache-2.0
def handle_nans(previous_state, next_state, step_size, step_size_max, kinetic_change): """if there are nans, let's reduce the stepsize, and not update the state. The function returns the old state in this case.""" reduced_step_size = 0.8 p, unravel_fn = ravel_pytree(next_state.position) nonans = jn...
if there are nans, let's reduce the stepsize, and not update the state. The function returns the old state in this case.
handle_nans
python
blackjax-devs/blackjax
blackjax/adaptation/adjusted_mclmc_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/adjusted_mclmc_adaptation.py
Apache-2.0
def get_filter_adapt_info_fn( state_keys: Set[str] = set(), info_keys: Set[str] = set(), adapt_state_keys: Set[str] = set(), ): """Generate a function to filter what is saved in AdaptationInfo. Used for adptation_info_fn parameters of the adaptation algorithms. adaptation_info_fn=get_filter_ada...
Generate a function to filter what is saved in AdaptationInfo. Used for adptation_info_fn parameters of the adaptation algorithms. adaptation_info_fn=get_filter_adapt_info_fn() saves no auxiliary information
get_filter_adapt_info_fn
python
blackjax-devs/blackjax
blackjax/adaptation/base.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/base.py
Apache-2.0
def base( jitter_generator: Callable, next_random_arg_fn: Callable, optim: optax.GradientTransformation, target_acceptance_rate: float, decay_rate: float, ) -> Tuple[Callable, Callable]: """Maximizing the Change in the Estimator of the Expected Square criterion (trajectory length) and dual a...
Maximizing the Change in the Estimator of the Expected Square criterion (trajectory length) and dual averaging procedure (step size) for the jittered Hamiltonian Monte Carlo kernel :cite:p:`hoffman2021adaptive`. This adaptation algorithm tunes the step size and trajectory length, i.e. number of integra...
base
python
blackjax-devs/blackjax
blackjax/adaptation/chees_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/chees_adaptation.py
Apache-2.0
def compute_parameters( proposed_positions: ArrayLikeTree, proposed_momentums: ArrayLikeTree, initial_positions: ArrayLikeTree, acceptance_probabilities: Array, is_divergent: Array, initial_adaptation_state: ChEESAdaptationState, ) -> ChEESAdaptationState: """...
Compute values for the parameters based on statistics collected from multiple chains. Parameters ---------- proposed_positions: A PyTree that contains the position proposed by the HMC algorithm of every chain (proposal that is accepted or rejected using MH). ...
compute_parameters
python
blackjax-devs/blackjax
blackjax/adaptation/chees_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/chees_adaptation.py
Apache-2.0
def update( adaptation_state: ChEESAdaptationState, proposed_positions: ArrayLikeTree, proposed_momentums: ArrayLikeTree, initial_positions: ArrayLikeTree, acceptance_probabilities: Array, is_divergent: Array, ): """Update the adaptation state and parameter va...
Update the adaptation state and parameter values. Parameters ---------- adaptation_state The current state of the adaptation algorithm proposed_positions: The position proposed by the HMC algorithm of every chain. proposed_momentums: The momen...
update
python
blackjax-devs/blackjax
blackjax/adaptation/chees_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/chees_adaptation.py
Apache-2.0
def chees_adaptation( logdensity_fn: Callable, num_chains: int, *, jitter_generator: Optional[Callable] = None, jitter_amount: float = 1.0, target_acceptance_rate: float = OPTIMAL_TARGET_ACCEPTANCE_RATE, decay_rate: float = 0.5, adaptation_info_fn: Callable = return_all_adapt_info, ) -> ...
Adapt the step size and trajectory length (number of integration steps / step size) parameters of the jittered HMC algorthm. The jittered HMC algorithm depends on the value of a step size, controlling the discretization step of the integrator, and a trajectory length, given by the number of integration...
chees_adaptation
python
blackjax-devs/blackjax
blackjax/adaptation/chees_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/chees_adaptation.py
Apache-2.0
def mass_matrix_adaptation( is_diagonal_matrix: bool = True, ) -> tuple[Callable, Callable, Callable]: """Adapts the values in the mass matrix by computing the covariance between parameters. Parameters ---------- is_diagonal_matrix When True the algorithm adapts and returns a diagonal m...
Adapts the values in the mass matrix by computing the covariance between parameters. Parameters ---------- is_diagonal_matrix When True the algorithm adapts and returns a diagonal mass matrix (default), otherwise adaps and returns a dense mass matrix. Returns ------- init ...
mass_matrix_adaptation
python
blackjax-devs/blackjax
blackjax/adaptation/mass_matrix.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mass_matrix.py
Apache-2.0
def init(n_dims: int) -> MassMatrixAdaptationState: """Initialize the matrix adaptation. Parameters ---------- ndims The number of dimensions of the mass matrix, which corresponds to the number of dimensions of the chain position. """ if is_diago...
Initialize the matrix adaptation. Parameters ---------- ndims The number of dimensions of the mass matrix, which corresponds to the number of dimensions of the chain position.
init
python
blackjax-devs/blackjax
blackjax/adaptation/mass_matrix.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mass_matrix.py
Apache-2.0
def update( mm_state: MassMatrixAdaptationState, position: ArrayLike ) -> MassMatrixAdaptationState: """Update the algorithm's state. Parameters ---------- state: The current state of the mass matrix adapation. position: The current position o...
Update the algorithm's state. Parameters ---------- state: The current state of the mass matrix adapation. position: The current position of the chain.
update
python
blackjax-devs/blackjax
blackjax/adaptation/mass_matrix.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mass_matrix.py
Apache-2.0
def final(mm_state: MassMatrixAdaptationState) -> MassMatrixAdaptationState: """Final iteration of the mass matrix adaptation. In this step we compute the mass matrix from the covariance matrix computed by the Welford algorithm, and re-initialize the later. """ _, wc_state = mm...
Final iteration of the mass matrix adaptation. In this step we compute the mass matrix from the covariance matrix computed by the Welford algorithm, and re-initialize the later.
final
python
blackjax-devs/blackjax
blackjax/adaptation/mass_matrix.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mass_matrix.py
Apache-2.0
def welford_algorithm(is_diagonal_matrix: bool) -> tuple[Callable, Callable, Callable]: r"""Welford's online estimator of covariance. It is possible to compute the variance of a population of values in an on-line fashion to avoid storing intermediate results. The naive recurrence relations between the ...
Welford's online estimator of covariance. It is possible to compute the variance of a population of values in an on-line fashion to avoid storing intermediate results. The naive recurrence relations between the sample mean and variance at a step and the next are however not numerically stable. Wel...
welford_algorithm
python
blackjax-devs/blackjax
blackjax/adaptation/mass_matrix.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mass_matrix.py
Apache-2.0
def init(n_dims: int) -> WelfordAlgorithmState: """Initialize the covariance estimation. When the matrix is diagonal it is sufficient to work with an array that contains the diagonal value. Otherwise we need to work with the matrix in full. Parameters ---------- n_dims:...
Initialize the covariance estimation. When the matrix is diagonal it is sufficient to work with an array that contains the diagonal value. Otherwise we need to work with the matrix in full. Parameters ---------- n_dims: int The number of dimensions of the problem, w...
init
python
blackjax-devs/blackjax
blackjax/adaptation/mass_matrix.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mass_matrix.py
Apache-2.0
def update( wa_state: WelfordAlgorithmState, value: ArrayLike ) -> WelfordAlgorithmState: """Update the M2 matrix using the new value. Parameters ---------- wa_state: The current state of the Welford Algorithm value: Array, shape (1,) The new ...
Update the M2 matrix using the new value. Parameters ---------- wa_state: The current state of the Welford Algorithm value: Array, shape (1,) The new sample (typically position of the chain) used to update m2
update
python
blackjax-devs/blackjax
blackjax/adaptation/mass_matrix.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mass_matrix.py
Apache-2.0
def mclmc_find_L_and_step_size( mclmc_kernel, num_steps, state, rng_key, frac_tune1=0.1, frac_tune2=0.1, frac_tune3=0.1, desired_energy_var=5e-4, trust_in_estimate=1.5, num_effective_samples=150, diagonal_preconditioning=True, params=None, ): """ Finds the optimal...
Finds the optimal value of the parameters for the MCLMC algorithm. Parameters ---------- mclmc_kernel The kernel function used for the MCMC algorithm. num_steps The number of MCMC steps that will subsequently be run, after tuning. state The initial state of the MCMC alg...
mclmc_find_L_and_step_size
python
blackjax-devs/blackjax
blackjax/adaptation/mclmc_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mclmc_adaptation.py
Apache-2.0
def make_L_step_size_adaptation( kernel, dim, frac_tune1, frac_tune2, diagonal_preconditioning, desired_energy_var=1e-3, trust_in_estimate=1.5, num_effective_samples=150, ): """Adapts the stepsize and L of the MCLMC kernel. Designed for unadjusted MCLMC""" decay_rate = (num_effe...
Adapts the stepsize and L of the MCLMC kernel. Designed for unadjusted MCLMC
make_L_step_size_adaptation
python
blackjax-devs/blackjax
blackjax/adaptation/mclmc_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mclmc_adaptation.py
Apache-2.0
def predictor(previous_state, params, adaptive_state, rng_key): """does one step with the dynamics and updates the prediction for the optimal stepsize Designed for the unadjusted MCHMC""" time, x_average, step_size_max = adaptive_state rng_key, nan_key = jax.random.split(rng_key) ...
does one step with the dynamics and updates the prediction for the optimal stepsize Designed for the unadjusted MCHMC
predictor
python
blackjax-devs/blackjax
blackjax/adaptation/mclmc_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/mclmc_adaptation.py
Apache-2.0
def base(): """Maximum-Eigenvalue Adaptation of damping and step size for the generalized Hamiltonian Monte Carlo kernel :cite:p:`hoffman2022tuning`. This algorithm performs a cross-chain adaptation scheme for the generalized HMC algorithm that automatically selects values for the generalized HMC's ...
Maximum-Eigenvalue Adaptation of damping and step size for the generalized Hamiltonian Monte Carlo kernel :cite:p:`hoffman2022tuning`. This algorithm performs a cross-chain adaptation scheme for the generalized HMC algorithm that automatically selects values for the generalized HMC's tunable parameter...
base
python
blackjax-devs/blackjax
blackjax/adaptation/meads_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/meads_adaptation.py
Apache-2.0
def compute_parameters( positions: ArrayLikeTree, logdensity_grad: ArrayLikeTree, current_iteration: int ): """Compute values for the parameters based on statistics collected from multiple chains. Parameters ---------- positions: A PyTree that contains th...
Compute values for the parameters based on statistics collected from multiple chains. Parameters ---------- positions: A PyTree that contains the current position of every chains. logdensity_grad: A PyTree that contains the gradients of the logdensity ...
compute_parameters
python
blackjax-devs/blackjax
blackjax/adaptation/meads_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/meads_adaptation.py
Apache-2.0
def update( adaptation_state: MEADSAdaptationState, positions: ArrayLikeTree, logdensity_grad: ArrayLikeTree, ) -> MEADSAdaptationState: """Update the adaptation state and parameter values. We find new optimal values for the parameters of the generalized HMC kernel u...
Update the adaptation state and parameter values. We find new optimal values for the parameters of the generalized HMC kernel using heuristics based on the maximum eigenvalue of the covariance and gradient matrices given by an ensemble of chains. Parameters ---------- a...
update
python
blackjax-devs/blackjax
blackjax/adaptation/meads_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/meads_adaptation.py
Apache-2.0
def meads_adaptation( logdensity_fn: Callable, num_chains: int, adaptation_info_fn: Callable = return_all_adapt_info, ) -> AdaptationAlgorithm: """Adapt the parameters of the Generalized HMC algorithm. The Generalized HMC algorithm depends on three parameters, each controlling one element of it...
Adapt the parameters of the Generalized HMC algorithm. The Generalized HMC algorithm depends on three parameters, each controlling one element of its behaviour: step size controls the integrator's dynamics, alpha controls the persistency of the momentum variable, and delta controls the deterministic tr...
meads_adaptation
python
blackjax-devs/blackjax
blackjax/adaptation/meads_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/meads_adaptation.py
Apache-2.0
def maximum_eigenvalue(matrix: ArrayLikeTree) -> Array: """Estimate the largest eigenvalues of a matrix. We calculate an unbiased estimate of the ratio between the sum of the squared eigenvalues and the sum of the eigenvalues from the input matrix. This ratio approximates the largest eigenvalue well ex...
Estimate the largest eigenvalues of a matrix. We calculate an unbiased estimate of the ratio between the sum of the squared eigenvalues and the sum of the eigenvalues from the input matrix. This ratio approximates the largest eigenvalue well except in cases when there are a large number of small eigenv...
maximum_eigenvalue
python
blackjax-devs/blackjax
blackjax/adaptation/meads_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/meads_adaptation.py
Apache-2.0
def base( target_acceptance_rate: float = 0.80, ): """Warmup scheme for sampling procedures based on euclidean manifold HMC. This adaptation runs in two steps: 1. The Pathfinder algorithm is ran and we subsequently compute an estimate for the value of the inverse mass matrix, as well as a new init...
Warmup scheme for sampling procedures based on euclidean manifold HMC. This adaptation runs in two steps: 1. The Pathfinder algorithm is ran and we subsequently compute an estimate for the value of the inverse mass matrix, as well as a new initialization point for the markov chain that is supposedly c...
base
python
blackjax-devs/blackjax
blackjax/adaptation/pathfinder_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/pathfinder_adaptation.py
Apache-2.0
def init( alpha, beta, gamma, initial_step_size: float, ) -> PathfinderAdaptationState: """Initialze the adaptation state and parameter values. We use the Pathfinder algorithm to compute an estimate of the inverse mass matrix that will stay constant throughou...
Initialze the adaptation state and parameter values. We use the Pathfinder algorithm to compute an estimate of the inverse mass matrix that will stay constant throughout the rest of the adaptation. Parameters ---------- alpha, beta, gamma Factored representa...
init
python
blackjax-devs/blackjax
blackjax/adaptation/pathfinder_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/pathfinder_adaptation.py
Apache-2.0
def update( adaptation_state: PathfinderAdaptationState, position: ArrayLikeTree, acceptance_rate: float, ) -> PathfinderAdaptationState: """Update the adaptation state and parameter values. Since the value of the inverse mass matrix is already known we only update t...
Update the adaptation state and parameter values. Since the value of the inverse mass matrix is already known we only update the state of the step size adaptation algorithm. Parameters ---------- adaptation_state Current adptation state. position ...
update
python
blackjax-devs/blackjax
blackjax/adaptation/pathfinder_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/pathfinder_adaptation.py
Apache-2.0
def final(warmup_state: PathfinderAdaptationState) -> tuple[float, Array]: """Return the final values for the step size and inverse mass matrix.""" step_size = jnp.exp(warmup_state.ss_state.log_step_size_avg) inverse_mass_matrix = warmup_state.inverse_mass_matrix return step_size, invers...
Return the final values for the step size and inverse mass matrix.
final
python
blackjax-devs/blackjax
blackjax/adaptation/pathfinder_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/pathfinder_adaptation.py
Apache-2.0
def pathfinder_adaptation( algorithm, logdensity_fn: Callable, initial_step_size: float = 1.0, target_acceptance_rate: float = 0.80, adaptation_info_fn: Callable = return_all_adapt_info, **extra_parameters, ) -> AdaptationAlgorithm: """Adapt the value of the inverse mass matrix and step size...
Adapt the value of the inverse mass matrix and step size parameters of algorithms in the HMC fmaily. Parameters ---------- algorithm The algorithm whose parameters are being tuned. logdensity_fn The log density probability density function from which we wish to sample. initial_s...
pathfinder_adaptation
python
blackjax-devs/blackjax
blackjax/adaptation/pathfinder_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/pathfinder_adaptation.py
Apache-2.0
def dual_averaging_adaptation( target: float, t0: int = 10, gamma: float = 0.05, kappa: float = 0.75 ) -> tuple[Callable, Callable, Callable]: """Tune the step size in order to achieve a desired target acceptance rate. Let us note :math:`\\epsilon` the current step size, :math:`\\alpha_t` the metropoli...
Tune the step size in order to achieve a desired target acceptance rate. Let us note :math:`\epsilon` the current step size, :math:`\alpha_t` the metropolis acceptance rate at time :math:`t` and :math:`\delta` the desired aceptance rate. We define: .. math: H_t = \delta - \alpha_t the err...
dual_averaging_adaptation
python
blackjax-devs/blackjax
blackjax/adaptation/step_size.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/step_size.py
Apache-2.0
def update( da_state: DualAveragingAdaptationState, acceptance_rate: float ) -> DualAveragingAdaptationState: """Update the state of the Dual Averaging adaptive algorithm. Parameters ---------- da_state: The current state of the dual averaging algorithm. ...
Update the state of the Dual Averaging adaptive algorithm. Parameters ---------- da_state: The current state of the dual averaging algorithm. acceptance_rate: float in [0, 1] The current metropolis acceptance rate. Returns ------- The upd...
update
python
blackjax-devs/blackjax
blackjax/adaptation/step_size.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/step_size.py
Apache-2.0
def find_reasonable_step_size( rng_key: PRNGKey, kernel_generator: Callable[[float], Callable], reference_state: HMCState, initial_step_size: float, target_accept: float = 0.65, ) -> float: """Find a reasonable initial step size during warmup. While the dual averaging scheme is guaranteed t...
Find a reasonable initial step size during warmup. While the dual averaging scheme is guaranteed to converge to a reasonable value for the step size starting from any value, choosing a good first value can speed up the convergence. This heuristics doubles and halves the step size until the acceptance p...
find_reasonable_step_size
python
blackjax-devs/blackjax
blackjax/adaptation/step_size.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/step_size.py
Apache-2.0
def do_continue(rss_state: ReasonableStepSizeState) -> bool: """Decides whether the search should continue. The search stops when it crosses the `target_accept` threshold, i.e. when the current direction is opposite to the previous direction. Note ---- Per JAX's documen...
Decides whether the search should continue. The search stops when it crosses the `target_accept` threshold, i.e. when the current direction is opposite to the previous direction. Note ---- Per JAX's documentation :cite:p:`jax_finfo` the `jnp.finfo` object is cached so we do not...
do_continue
python
blackjax-devs/blackjax
blackjax/adaptation/step_size.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/step_size.py
Apache-2.0
def update(rss_state: ReasonableStepSizeState) -> ReasonableStepSizeState: """Perform one step of the step size search.""" i, direction, _, step_size = rss_state subkey = jax.random.fold_in(rng_key, i) step_size = (2.0**direction) * step_size kernel = kernel_generator(step_size)...
Perform one step of the step size search.
update
python
blackjax-devs/blackjax
blackjax/adaptation/step_size.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/step_size.py
Apache-2.0
def base( is_mass_matrix_diagonal: bool, target_acceptance_rate: float = 0.80, ) -> tuple[Callable, Callable, Callable]: """Warmup scheme for sampling procedures based on euclidean manifold HMC. The schedule and algorithms used match Stan's :cite:p:`stan_hmc_param` as closely as possible. Unlike se...
Warmup scheme for sampling procedures based on euclidean manifold HMC. The schedule and algorithms used match Stan's :cite:p:`stan_hmc_param` as closely as possible. Unlike several other libraries, we separate the warmup and sampling phases explicitly. This ensure a better modularity; a change in the warmu...
base
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def init( position: ArrayLikeTree, initial_step_size: float ) -> WindowAdaptationState: """Initialze the adaptation state and parameter values. Unlike the original Stan window adaptation we do not use the `find_reasonable_step_size` algorithm which we found to be unnecessary. ...
Initialze the adaptation state and parameter values. Unlike the original Stan window adaptation we do not use the `find_reasonable_step_size` algorithm which we found to be unnecessary. We may reconsider this choice in the future.
init
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def fast_update( position: ArrayLikeTree, acceptance_rate: float, warmup_state: WindowAdaptationState, ) -> WindowAdaptationState: """Update the adaptation state when in a "fast" window. Only the step size is adapted in fast windows. "Fast" refers to the fact that th...
Update the adaptation state when in a "fast" window. Only the step size is adapted in fast windows. "Fast" refers to the fact that the optimization algorithms are relatively fast to converge compared to the covariance estimation with Welford's algorithm
fast_update
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def slow_update( position: ArrayLikeTree, acceptance_rate: float, warmup_state: WindowAdaptationState, ) -> WindowAdaptationState: """Update the adaptation state when in a "slow" window. Both the mass matrix adaptation *state* and the step size state are adapted in s...
Update the adaptation state when in a "slow" window. Both the mass matrix adaptation *state* and the step size state are adapted in slow windows. The value of the step size is updated as well, but the new value of the inverse mass matrix is only computed at the end of the slow window. "...
slow_update
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def slow_final(warmup_state: WindowAdaptationState) -> WindowAdaptationState: """Update the parameters at the end of a slow adaptation window. We compute the value of the mass matrix and reset the mass matrix adapation's internal state since middle windows are "memoryless". """ ...
Update the parameters at the end of a slow adaptation window. We compute the value of the mass matrix and reset the mass matrix adapation's internal state since middle windows are "memoryless".
slow_final
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def update( adaptation_state: WindowAdaptationState, adaptation_stage: tuple, position: ArrayLikeTree, acceptance_rate: float, ) -> WindowAdaptationState: """Update the adaptation state and parameter values. Parameters ---------- adaptation_state ...
Update the adaptation state and parameter values. Parameters ---------- adaptation_state Current adptation state. adaptation_stage The current stage of the warmup: whether this is a slow window, a fast window and if we are at the last step of a slow w...
update
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def final(warmup_state: WindowAdaptationState) -> tuple[float, Array]: """Return the final values for the step size and mass matrix.""" step_size = jnp.exp(warmup_state.ss_state.log_step_size_avg) inverse_mass_matrix = warmup_state.imm_state.inverse_mass_matrix return step_size, inverse_...
Return the final values for the step size and mass matrix.
final
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def window_adaptation( algorithm, logdensity_fn: Callable, is_mass_matrix_diagonal: bool = True, initial_step_size: float = 1.0, target_acceptance_rate: float = 0.80, progress_bar: bool = False, adaptation_info_fn: Callable = return_all_adapt_info, integrator=mcmc.integrators.velocity_ve...
Adapt the value of the inverse mass matrix and step size parameters of algorithms in the HMC fmaily. See Blackjax.hmc_family Algorithms in the HMC family on a euclidean manifold depend on the value of at least two parameters: the step size, related to the trajectory integrator, and the mass matrix, lin...
window_adaptation
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def build_schedule( num_steps: int, initial_buffer_size: int = 75, final_buffer_size: int = 50, first_window_size: int = 25, ) -> list[tuple[int, bool]]: """Return the schedule for Stan's warmup. The schedule below is intended to be as close as possible to Stan's :cite:p:`stan_hmc_param`. T...
Return the schedule for Stan's warmup. The schedule below is intended to be as close as possible to Stan's :cite:p:`stan_hmc_param`. The warmup period is split into three stages: 1. An initial fast interval to reach the typical set. Only the step size is adapted in this window. 2. "Slow" parameter...
build_schedule
python
blackjax-devs/blackjax
blackjax/adaptation/window_adaptation.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/adaptation/window_adaptation.py
Apache-2.0
def build_kernel( logdensity_fn: Callable, integrator: Callable = integrators.isokinetic_mclachlan, divergence_threshold: float = 1000, inverse_mass_matrix=1.0, ): """Build an MHMCHMC kernel where the number of integration steps is chosen randomly. Parameters ---------- integrator ...
Build an MHMCHMC kernel where the number of integration steps is chosen randomly. Parameters ---------- integrator The integrator to use to integrate the Hamiltonian dynamics. divergence_threshold Value of the difference in energy above which we consider that the transition is divergent...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/adjusted_mclmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/adjusted_mclmc.py
Apache-2.0
def kernel( rng_key: PRNGKey, state: HMCState, step_size: float, num_integration_steps: int, L_proposal_factor: float = jnp.inf, ) -> tuple[HMCState, HMCInfo]: """Generate a new sample with the MHMCHMC kernel.""" key_momentum, key_integrator = jax.random.spli...
Generate a new sample with the MHMCHMC kernel.
kernel
python
blackjax-devs/blackjax
blackjax/mcmc/adjusted_mclmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/adjusted_mclmc.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, step_size: float, L_proposal_factor: float = jnp.inf, inverse_mass_matrix=1.0, *, divergence_threshold: int = 1000, integrator: Callable = integrators.isokinetic_mclachlan, num_integration_steps, ) -> SamplingAlgorithm: """Implements the...
Implements the (basic) user interface for the MHMCHMC kernel. Parameters ---------- logdensity_fn The log-density function we wish to draw samples from. step_size The value to use for the step size in the symplectic integrator. divergence_threshold The absolute value of the ...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/adjusted_mclmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/adjusted_mclmc.py
Apache-2.0
def adjusted_mclmc_proposal( integrator: Callable, step_size: Union[float, ArrayLikeTree], L_proposal_factor: float, num_integration_steps: int = 1, divergence_threshold: float = 1000, *, sample_proposal: Callable = static_binomial_sampling, ) -> Callable: """Vanilla MHMCHMC algorithm. ...
Vanilla MHMCHMC algorithm. The algorithm integrates the trajectory applying a integrator `num_integration_steps` times in one direction to get a proposal and uses a Metropolis-Hastings acceptance step to either reject or accept this proposal. This is what people usually refer to when they talk about "t...
adjusted_mclmc_proposal
python
blackjax-devs/blackjax
blackjax/mcmc/adjusted_mclmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/adjusted_mclmc.py
Apache-2.0
def build_kernel( integration_steps_fn, integrator: Callable = integrators.isokinetic_mclachlan, divergence_threshold: float = 1000, next_random_arg_fn: Callable = lambda key: jax.random.split(key)[1], inverse_mass_matrix=1.0, ): """Build a Dynamic MHMCHMC kernel where the number of integration ...
Build a Dynamic MHMCHMC kernel where the number of integration steps is chosen randomly. Parameters ---------- integrator The integrator to use to integrate the Hamiltonian dynamics. divergence_threshold Value of the difference in energy above which we consider that the transition is di...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/adjusted_mclmc_dynamic.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/adjusted_mclmc_dynamic.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, step_size: float, L_proposal_factor: float = jnp.inf, inverse_mass_matrix=1.0, *, divergence_threshold: int = 1000, integrator: Callable = integrators.isokinetic_mclachlan, next_random_arg_fn: Callable = lambda key: jax.random.split(key)[1],...
Implements the (basic) user interface for the dynamic MHMCHMC kernel. Parameters ---------- logdensity_fn The log-density function we wish to draw samples from. step_size The value to use for the step size in the symplectic integrator. divergence_threshold The absolute value...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/adjusted_mclmc_dynamic.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/adjusted_mclmc_dynamic.py
Apache-2.0
def rescale(mu): """returns s, such that round(U(0, 1) * s + 0.5) has expected value mu. """ k = jnp.floor(2 * mu - 1) x = k * (mu - 0.5 * (k + 1)) / (k + 1 - mu) return k + x
returns s, such that round(U(0, 1) * s + 0.5) has expected value mu.
rescale
python
blackjax-devs/blackjax
blackjax/mcmc/adjusted_mclmc_dynamic.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/adjusted_mclmc_dynamic.py
Apache-2.0
def build_kernel(): """Build a Barker's proposal kernel. Returns ------- A kernel that takes a rng_key and a Pytree that contains the current state of the chain and that returns a new state of the chain along with information about the transition. """ def _compute_acceptance_probabili...
Build a Barker's proposal kernel. Returns ------- A kernel that takes a rng_key and a Pytree that contains the current state of the chain and that returns a new state of the chain along with information about the transition.
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/barker.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/barker.py
Apache-2.0
def _compute_acceptance_probability( state: BarkerState, proposal: BarkerState, metric: Metric ) -> Numeric: """Compute the acceptance probability of the Barker's proposal kernel.""" x = state.position y = proposal.position log_x = state.logdensity_grad log_y = propo...
Compute the acceptance probability of the Barker's proposal kernel.
_compute_acceptance_probability
python
blackjax-devs/blackjax
blackjax/mcmc/barker.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/barker.py
Apache-2.0
def kernel( rng_key: PRNGKey, state: BarkerState, logdensity_fn: Callable, step_size: float, inverse_mass_matrix: metrics.MetricTypes | None = None, ) -> tuple[BarkerState, BarkerInfo]: """Generate a new sample with the Barker kernel.""" if inverse_mass_matrix...
Generate a new sample with the Barker kernel.
kernel
python
blackjax-devs/blackjax
blackjax/mcmc/barker.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/barker.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, step_size: float, inverse_mass_matrix: metrics.MetricTypes | None = None, ) -> SamplingAlgorithm: """Implements the (basic) user interface for the Barker's proposal :cite:p:`Livingstone2022Barker` kernel with a Gaussian base kernel. The general Bar...
Implements the (basic) user interface for the Barker's proposal :cite:p:`Livingstone2022Barker` kernel with a Gaussian base kernel. The general Barker kernel builder (:meth:`blackjax.mcmc.barker.build_kernel`, alias `blackjax.barker.build_kernel`) can be cumbersome to manipulate. Since most users only need...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/barker.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/barker.py
Apache-2.0
def _barker_sample(key, mean, a, scale, metric): r""" Sample from a multivariate Barker's proposal distribution for PyTrees. Parameters ---------- key A PRNG key. mean The mean of the normal distribution, a PyTree. This corresponds to :math:`\mu` in the equation above. a ...
Sample from a multivariate Barker's proposal distribution for PyTrees. Parameters ---------- key A PRNG key. mean The mean of the normal distribution, a PyTree. This corresponds to :math:`\mu` in the equation above. a The parameter :math:`a` in the equation above, the s...
_barker_sample
python
blackjax-devs/blackjax
blackjax/mcmc/barker.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/barker.py
Apache-2.0
def overdamped_langevin(logdensity_grad_fn): """Euler solver for overdamped Langevin diffusion.""" def one_step(rng_key, state: DiffusionState, step_size: float, batch: tuple = ()): position, _, logdensity_grad = state noise = generate_gaussian_noise(rng_key, position) position = jax.tr...
Euler solver for overdamped Langevin diffusion.
overdamped_langevin
python
blackjax-devs/blackjax
blackjax/mcmc/diffusions.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/diffusions.py
Apache-2.0
def build_kernel( integrator: Callable = integrators.velocity_verlet, divergence_threshold: float = 1000, next_random_arg_fn: Callable = lambda key: jax.random.split(key)[1], integration_steps_fn: Callable = lambda key: jax.random.randint(key, (), 1, 10), ): """Build a Dynamic HMC kernel where the n...
Build a Dynamic HMC kernel where the number of integration steps is chosen randomly. Parameters ---------- integrator The symplectic integrator to use to integrate the Hamiltonian dynamics. divergence_threshold Value of the difference in energy above which we consider that the transitio...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/dynamic_hmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/dynamic_hmc.py
Apache-2.0
def kernel( rng_key: PRNGKey, state: DynamicHMCState, logdensity_fn: Callable, step_size: float, inverse_mass_matrix: Array, **integration_steps_kwargs, ) -> tuple[DynamicHMCState, HMCInfo]: """Generate a new sample with the HMC kernel.""" num_integrat...
Generate a new sample with the HMC kernel.
kernel
python
blackjax-devs/blackjax
blackjax/mcmc/dynamic_hmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/dynamic_hmc.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, step_size: float, inverse_mass_matrix: Array, *, divergence_threshold: int = 1000, integrator: Callable = integrators.velocity_verlet, next_random_arg_fn: Callable = lambda key: jax.random.split(key)[1], integration_steps_fn: Callable = lamb...
Implements the (basic) user interface for the dynamic HMC kernel. Parameters ---------- logdensity_fn The log-density function we wish to draw samples from. step_size The value to use for the step size in the symplectic integrator. inverse_mass_matrix The value to use for th...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/dynamic_hmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/dynamic_hmc.py
Apache-2.0
def halton_trajectory_length( i: Array, trajectory_length_adjustment: float, max_bits: int = 10 ) -> int: """Generate a quasi-random number of integration steps.""" s = rescale(trajectory_length_adjustment) return jnp.asarray(jnp.rint(0.5 + halton_sequence(i, max_bits) * s), dtype=int)
Generate a quasi-random number of integration steps.
halton_trajectory_length
python
blackjax-devs/blackjax
blackjax/mcmc/dynamic_hmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/dynamic_hmc.py
Apache-2.0
def build_kernel(cov_matrix: Array, mean: Array): """Build an Elliptical Slice sampling kernel :cite:p:`murray2010elliptical`. Parameters ---------- cov_matrix The value of the covariance matrix of the gaussian prior distribution from the posterior we wish to sample. Returns --...
Build an Elliptical Slice sampling kernel :cite:p:`murray2010elliptical`. Parameters ---------- cov_matrix The value of the covariance matrix of the gaussian prior distribution from the posterior we wish to sample. Returns ------- A kernel that takes a rng_key and a Pytree that...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/elliptical_slice.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/elliptical_slice.py
Apache-2.0
def as_top_level_api( loglikelihood_fn: Callable, *, mean: Array, cov: Array, ) -> SamplingAlgorithm: """Implements the (basic) user interface for the Elliptical Slice sampling kernel. Examples -------- A new Elliptical Slice sampling kernel can be initialized and used with the followi...
Implements the (basic) user interface for the Elliptical Slice sampling kernel. Examples -------- A new Elliptical Slice sampling kernel can be initialized and used with the following code: .. code:: ellip_slice = blackjax.elliptical_slice(loglikelihood_fn, cov_matrix) state = ellip_...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/elliptical_slice.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/elliptical_slice.py
Apache-2.0
def elliptical_proposal( logdensity_fn: Callable, momentum_generator: Callable, mean: Array, ) -> Callable: """Build an Ellitpical slice sampling kernel. The algorithm samples a latent parameter, traces an ellipse connecting the initial position and the latent parameter and does slice sampling ...
Build an Ellitpical slice sampling kernel. The algorithm samples a latent parameter, traces an ellipse connecting the initial position and the latent parameter and does slice sampling on this ellipse to output a new sample from the posterior distribution. Parameters ---------- logdensity_fn ...
elliptical_proposal
python
blackjax-devs/blackjax
blackjax/mcmc/elliptical_slice.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/elliptical_slice.py
Apache-2.0
def slice_fn(vals): """Perform slice sampling around the ellipsis. Checks if the proposed position's likelihood is larger than the slice variable. Returns the position if True, shrinks the bracket for sampling `theta` and samples a new proposal if False. As ...
Perform slice sampling around the ellipsis. Checks if the proposed position's likelihood is larger than the slice variable. Returns the position if True, shrinks the bracket for sampling `theta` and samples a new proposal if False. As the bracket `[theta_min, theta_max]...
slice_fn
python
blackjax-devs/blackjax
blackjax/mcmc/elliptical_slice.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/elliptical_slice.py
Apache-2.0
def ellipsis(position, momentum, theta, mean): """Generate proposal from the ellipsis. Given a scalar theta indicating a point on the circumference of the ellipsis and the shared mean vector for both position and momentum variables, generate proposed position and momentum to later accept or reject ...
Generate proposal from the ellipsis. Given a scalar theta indicating a point on the circumference of the ellipsis and the shared mean vector for both position and momentum variables, generate proposed position and momentum to later accept or reject depending on the slice variable.
ellipsis
python
blackjax-devs/blackjax
blackjax/mcmc/elliptical_slice.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/elliptical_slice.py
Apache-2.0
def build_kernel( noise_fn: Callable = lambda _: 0.0, divergence_threshold: float = 1000, ): """Build a Generalized HMC kernel. The Generalized HMC kernel performs a similar procedure to the standard HMC kernel with the difference of a persistent momentum variable and a non-reversible Metropoli...
Build a Generalized HMC kernel. The Generalized HMC kernel performs a similar procedure to the standard HMC kernel with the difference of a persistent momentum variable and a non-reversible Metropolis-Hastings step instead of the standard Metropolis-Hastings acceptance step. This means that; apart from...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/ghmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/ghmc.py
Apache-2.0
def kernel( rng_key: PRNGKey, state: GHMCState, logdensity_fn: Callable, step_size: float, momentum_inverse_scale: ArrayLikeTree, alpha: float, delta: float, ) -> tuple[GHMCState, hmc.HMCInfo]: """Generate new sample with the Generalized HMC kernel. ...
Generate new sample with the Generalized HMC kernel. Parameters ---------- rng_key JAX's pseudo random number generating key. state Current state of the chain. logdensity_fn (Unnormalized) Log density function being targeted. step_size...
kernel
python
blackjax-devs/blackjax
blackjax/mcmc/ghmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/ghmc.py
Apache-2.0
def update_momentum(rng_key, state, alpha, momentum_generator): """Persistent update of the momentum variable. Performs a persistent update of the momentum, taking as input the previous momentum, a random number generating key, the parameter alpha and the momentum generator function. Outputs an upd...
Persistent update of the momentum variable. Performs a persistent update of the momentum, taking as input the previous momentum, a random number generating key, the parameter alpha and the momentum generator function. Outputs an updated momentum that is a mixture of the previous momentum a new sample ...
update_momentum
python
blackjax-devs/blackjax
blackjax/mcmc/ghmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/ghmc.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, step_size: float, momentum_inverse_scale: ArrayLikeTree, alpha: float, delta: float, *, divergence_threshold: int = 1000, noise_gn: Callable = lambda _: 0.0, ) -> SamplingAlgorithm: """Implements the (basic) user interface for the Genera...
Implements the (basic) user interface for the Generalized HMC kernel. The Generalized HMC kernel performs a similar procedure to the standard HMC kernel with the difference of a persistent momentum variable and a non-reversible Metropolis-Hastings step instead of the standard Metropolis-Hastings acceptance...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/ghmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/ghmc.py
Apache-2.0
def build_kernel( integrator: Callable = integrators.velocity_verlet, divergence_threshold: float = 1000, ): """Build a HMC kernel. Parameters ---------- integrator The symplectic integrator to use to integrate the Hamiltonian dynamics. divergence_threshold Value of the diff...
Build a HMC kernel. Parameters ---------- integrator The symplectic integrator to use to integrate the Hamiltonian dynamics. divergence_threshold Value of the difference in energy above which we consider that the transition is divergent. Returns ------- A kernel tha...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/hmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/hmc.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, step_size: float, inverse_mass_matrix: metrics.MetricTypes, num_integration_steps: int, *, divergence_threshold: int = 1000, integrator: Callable = integrators.velocity_verlet, ) -> SamplingAlgorithm: """Implements the (basic) user interface...
Implements the (basic) user interface for the HMC kernel. The general hmc kernel builder (:meth:`blackjax.mcmc.hmc.build_kernel`, alias `blackjax.hmc.build_kernel`) can be cumbersome to manipulate. Since most users only need to specify the kernel parameters at initialization time, we provide a helper f...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/hmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/hmc.py
Apache-2.0
def hmc_proposal( integrator: Callable, kinetic_energy: metrics.KineticEnergy, step_size: Union[float, ArrayLikeTree], num_integration_steps: int = 1, divergence_threshold: float = 1000, *, sample_proposal: Callable = static_binomial_sampling, ) -> Callable: """Vanilla HMC algorithm. ...
Vanilla HMC algorithm. The algorithm integrates the trajectory applying a symplectic integrator `num_integration_steps` times in one direction to get a proposal and uses a Metropolis-Hastings acceptance step to either reject or accept this proposal. This is what people usually refer to when they talk a...
hmc_proposal
python
blackjax-devs/blackjax
blackjax/mcmc/hmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/hmc.py
Apache-2.0
def flip_momentum( state: integrators.IntegratorState, ) -> integrators.IntegratorState: """Flip the momentum at the end of the trajectory. To guarantee time-reversibility (hence detailed balance) we need to flip the last state's momentum. If we run the hamiltonian dynamics starting from the last s...
Flip the momentum at the end of the trajectory. To guarantee time-reversibility (hence detailed balance) we need to flip the last state's momentum. If we run the hamiltonian dynamics starting from the last state with flipped momentum we should indeed retrieve the initial state (with flipped momentum). ...
flip_momentum
python
blackjax-devs/blackjax
blackjax/mcmc/hmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/hmc.py
Apache-2.0
def generalized_two_stage_integrator( operator1: Callable, operator2: Callable, coefficients: list[float], format_output_fn: Callable = lambda x: x, ): """Generalized numerical integrator for solving ODEs. The generalized integrator performs numerical integration of a ODE system by alernati...
Generalized numerical integrator for solving ODEs. The generalized integrator performs numerical integration of a ODE system by alernating between stage 1 and stage 2 updates. The update scheme is decided by the coefficients, The scheme should be palindromic, i.e. the coefficients of the update scheme ...
generalized_two_stage_integrator
python
blackjax-devs/blackjax
blackjax/mcmc/integrators.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/integrators.py
Apache-2.0
def generate_euclidean_integrator(coefficients): """Generate symplectic integrator for solving a Hamiltonian system. The resulting integrator is volume-preserve and preserves the symplectic structure of phase space. """ def euclidean_integrator( logdensity_fn: Callable, kinetic_energy_fn: ...
Generate symplectic integrator for solving a Hamiltonian system. The resulting integrator is volume-preserve and preserves the symplectic structure of phase space.
generate_euclidean_integrator
python
blackjax-devs/blackjax
blackjax/mcmc/integrators.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/integrators.py
Apache-2.0
def update( momentum: ArrayTree, logdensity_grad: ArrayTree, step_size: float, coef: float, previous_kinetic_energy_change=None, is_last_call=False, ): """Momentum update based on Esh dynamics. The momentum updating map of the esh dynamics as derived ...
Momentum update based on Esh dynamics. The momentum updating map of the esh dynamics as derived in :cite:p:`steeg2021hamiltonian` There are no exponentials e^delta, which prevents overflows when the gradient norm is large.
update
python
blackjax-devs/blackjax
blackjax/mcmc/integrators.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/integrators.py
Apache-2.0
def partially_refresh_momentum(momentum, rng_key, step_size, L): """Adds a small noise to momentum and normalizes. Parameters ---------- rng_key The pseudo-random number generator key used to generate random numbers. momentum PyTree that the structure the output should to match. ...
Adds a small noise to momentum and normalizes. Parameters ---------- rng_key The pseudo-random number generator key used to generate random numbers. momentum PyTree that the structure the output should to match. step_size Step size L controls rate of momentum cha...
partially_refresh_momentum
python
blackjax-devs/blackjax
blackjax/mcmc/integrators.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/integrators.py
Apache-2.0
def solve_fixed_point_iteration( func: Callable[[ArrayTree], Tuple[ArrayTree, ArrayTree]], x0: ArrayTree, *, convergence_tol: float = 1e-6, divergence_tol: float = 1e10, max_iters: int = 100, norm_fn: Callable[[ArrayTree], float] = lambda x: jnp.max(jnp.abs(x)), ) -> Tuple[ArrayTree, ArrayTr...
Solve for x = func(x) using a fixed point iteration
solve_fixed_point_iteration
python
blackjax-devs/blackjax
blackjax/mcmc/integrators.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/integrators.py
Apache-2.0
def implicit_midpoint( logdensity_fn: Callable, kinetic_energy_fn: KineticEnergy, *, solver: FixedPointSolver = solve_fixed_point_iteration, **solver_kwargs: Any, ) -> Integrator: """The implicit midpoint integrator with support for non-stationary kinetic energy This is an integrator based ...
The implicit midpoint integrator with support for non-stationary kinetic energy This is an integrator based on :cite:t:`brofos2021evaluating`, which provides support for kinetic energies that depend on position. This integrator requires that the kinetic energy function takes two arguments: position and mom...
implicit_midpoint
python
blackjax-devs/blackjax
blackjax/mcmc/integrators.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/integrators.py
Apache-2.0
def build_kernel(): """Build a MALA kernel. Returns ------- A kernel that takes a rng_key and a Pytree that contains the current state of the chain and that returns a new state of the chain along with information about the transition. """ def transition_energy(state, new_state, step_s...
Build a MALA kernel. Returns ------- A kernel that takes a rng_key and a Pytree that contains the current state of the chain and that returns a new state of the chain along with information about the transition.
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/mala.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/mala.py
Apache-2.0
def transition_energy(state, new_state, step_size): """Transition energy to go from `state` to `new_state`""" theta = jax.tree_util.tree_map( lambda x, new_x, g: x - new_x - step_size * g, state.position, new_state.position, new_state.logdensity_grad, ...
Transition energy to go from `state` to `new_state`
transition_energy
python
blackjax-devs/blackjax
blackjax/mcmc/mala.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/mala.py
Apache-2.0
def kernel( rng_key: PRNGKey, state: MALAState, logdensity_fn: Callable, step_size: float ) -> tuple[MALAState, MALAInfo]: """Generate a new sample with the MALA kernel.""" grad_fn = jax.value_and_grad(logdensity_fn) integrator = diffusions.overdamped_langevin(grad_fn) key_i...
Generate a new sample with the MALA kernel.
kernel
python
blackjax-devs/blackjax
blackjax/mcmc/mala.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/mala.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, step_size: float, ) -> SamplingAlgorithm: """Implements the (basic) user interface for the MALA kernel. The general mala kernel builder (:meth:`blackjax.mcmc.mala.build_kernel`, alias `blackjax.mala.build_kernel`) can be cumbersome to manipulate. Since...
Implements the (basic) user interface for the MALA kernel. The general mala kernel builder (:meth:`blackjax.mcmc.mala.build_kernel`, alias `blackjax.mala.build_kernel`) can be cumbersome to manipulate. Since most users only need to specify the kernel parameters at initialization time, we provide a helper f...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/mala.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/mala.py
Apache-2.0
def svd_from_covariance(covariance: Array) -> CovarianceSVD: """Compute the singular value decomposition of the covariance matrix. Parameters ---------- covariance The covariance matrix. Returns ------- A ``CovarianceSVD`` object. """ U, Gamma, U_t = jnp.linalg.svd(covaria...
Compute the singular value decomposition of the covariance matrix. Parameters ---------- covariance The covariance matrix. Returns ------- A ``CovarianceSVD`` object.
svd_from_covariance
python
blackjax-devs/blackjax
blackjax/mcmc/marginal_latent_gaussian.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/marginal_latent_gaussian.py
Apache-2.0
def generate_mean_shifted_logprob(logdensity_fn, mean, covariance): """Generate a log-density function that is shifted by a constant Parameters ---------- logdensity_fn The original log-density function mean The mean of the prior Gaussian density covariance The covarianc...
Generate a log-density function that is shifted by a constant Parameters ---------- logdensity_fn The original log-density function mean The mean of the prior Gaussian density covariance The covariance of the prior Gaussian density. Returns ------- A log-density...
generate_mean_shifted_logprob
python
blackjax-devs/blackjax
blackjax/mcmc/marginal_latent_gaussian.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/marginal_latent_gaussian.py
Apache-2.0
def init(position, logdensity_fn, U_t): """Initialize the marginal version of the auxiliary gradient-based sampler. Parameters ---------- position The initial position of the chain. logdensity_fn The logarithm of the likelihood function for the latent Gaussian model. U_t ...
Initialize the marginal version of the auxiliary gradient-based sampler. Parameters ---------- position The initial position of the chain. logdensity_fn The logarithm of the likelihood function for the latent Gaussian model. U_t The unitary array of the covariance matrix. ...
init
python
blackjax-devs/blackjax
blackjax/mcmc/marginal_latent_gaussian.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/marginal_latent_gaussian.py
Apache-2.0
def build_kernel(cov_svd: CovarianceSVD): """Build the marginal version of the auxiliary gradient-based sampler. Parameters ---------- cov_svd The singular value decomposition of the covariance matrix. Returns ------- A kernel that takes a rng_key and a Pytree that contains the cur...
Build the marginal version of the auxiliary gradient-based sampler. Parameters ---------- cov_svd The singular value decomposition of the covariance matrix. Returns ------- A kernel that takes a rng_key and a Pytree that contains the current state of the chain and that returns a ne...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/marginal_latent_gaussian.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/marginal_latent_gaussian.py
Apache-2.0
def build_kernel( logdensity_fn, inverse_mass_matrix, integrator, desired_energy_var_max_ratio=jnp.inf, desired_energy_var=5e-4, ): """Build a HMC kernel. Parameters ---------- integrator The symplectic integrator to use to integrate the Hamiltonian dynamics. L t...
Build a HMC kernel. Parameters ---------- integrator The symplectic integrator to use to integrate the Hamiltonian dynamics. L the momentum decoherence rate. step_size step size of the integrator. Returns ------- A kernel that takes a rng_key and a Pytree that c...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/mclmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/mclmc.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, L, step_size, integrator=isokinetic_mclachlan, inverse_mass_matrix=1.0, desired_energy_var_max_ratio=jnp.inf, ) -> SamplingAlgorithm: """The general mclmc kernel builder (:meth:`blackjax.mcmc.mclmc.build_kernel`, alias `blackjax.mclmc.build_kern...
The general mclmc kernel builder (:meth:`blackjax.mcmc.mclmc.build_kernel`, alias `blackjax.mclmc.build_kernel`) can be cumbersome to manipulate. Since most users only need to specify the kernel parameters at initialization time, we provide a helper function that specializes the general kernel. We also...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/mclmc.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/mclmc.py
Apache-2.0
def default_metric(metric: MetricTypes) -> Metric: """Convert an input metric into a ``Metric`` object following sensible default rules The metric can be specified in three different ways: - A ``Metric`` object that implements the full interface - An ``Array`` which is assumed to specify the inverse m...
Convert an input metric into a ``Metric`` object following sensible default rules The metric can be specified in three different ways: - A ``Metric`` object that implements the full interface - An ``Array`` which is assumed to specify the inverse mass matrix of a static metric - A function that ...
default_metric
python
blackjax-devs/blackjax
blackjax/mcmc/metrics.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/metrics.py
Apache-2.0
def gaussian_euclidean( inverse_mass_matrix: Array, ) -> Metric: r"""Hamiltonian dynamic on euclidean manifold with normally-distributed momentum :cite:p:`betancourt2013general`. The gaussian euclidean metric is a euclidean metric further characterized by setting the conditional probability density...
Hamiltonian dynamic on euclidean manifold with normally-distributed momentum :cite:p:`betancourt2013general`. The gaussian euclidean metric is a euclidean metric further characterized by setting the conditional probability density :math:`\pi(momentum|position)` to follow a standard gaussian distributio...
gaussian_euclidean
python
blackjax-devs/blackjax
blackjax/mcmc/metrics.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/metrics.py
Apache-2.0
def is_turning( momentum_left: ArrayLikeTree, momentum_right: ArrayLikeTree, momentum_sum: ArrayLikeTree, position_left: Optional[ArrayLikeTree] = None, position_right: Optional[ArrayLikeTree] = None, ) -> bool: """Generalized U-turn criterion :cite:p:`betancourt2013g...
Generalized U-turn criterion :cite:p:`betancourt2013generalizing,nuts_uturn`. Parameters ---------- momentum_left Momentum of the leftmost point of the trajectory. momentum_right Momentum of the rightmost point of the trajectory. momentum_sum ...
is_turning
python
blackjax-devs/blackjax
blackjax/mcmc/metrics.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/metrics.py
Apache-2.0
def scale( position: ArrayLikeTree, element: ArrayLikeTree, *, inv: bool, trans: bool, ) -> ArrayLikeTree: """Scale elements by the mass matrix. Parameters ---------- position The current position. Not used in this metric. ...
Scale elements by the mass matrix. Parameters ---------- position The current position. Not used in this metric. elements Elements to scale inv Whether to scale the elements by the inverse mass matrix or the mass matrix. If True, t...
scale
python
blackjax-devs/blackjax
blackjax/mcmc/metrics.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/metrics.py
Apache-2.0
def scale( position: ArrayLikeTree, element: ArrayLikeTree, *, inv: bool, trans: bool, ) -> ArrayLikeTree: """Scale elements by the mass matrix. Parameters ---------- position The current position. Returns ------- ...
Scale elements by the mass matrix. Parameters ---------- position The current position. Returns ------- scaled_elements The scaled elements.
scale
python
blackjax-devs/blackjax
blackjax/mcmc/metrics.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/metrics.py
Apache-2.0
def build_kernel( integrator: Callable = integrators.velocity_verlet, divergence_threshold: int = 1000, ): """Build an iterative NUTS kernel. This algorithm is an iteration on the original NUTS algorithm :cite:p:`hoffman2014no` with two major differences: - We do not use slice samplig but mult...
Build an iterative NUTS kernel. This algorithm is an iteration on the original NUTS algorithm :cite:p:`hoffman2014no` with two major differences: - We do not use slice samplig but multinomial sampling for the proposal :cite:p:`betancourt2017conceptual`; - The trajectory expansion is not recursiv...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/nuts.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/nuts.py
Apache-2.0
def kernel( rng_key: PRNGKey, state: hmc.HMCState, logdensity_fn: Callable, step_size: float, inverse_mass_matrix: metrics.MetricTypes, max_num_doublings: int = 10, ) -> tuple[hmc.HMCState, NUTSInfo]: """Generate a new sample with the NUTS kernel.""" ...
Generate a new sample with the NUTS kernel.
kernel
python
blackjax-devs/blackjax
blackjax/mcmc/nuts.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/nuts.py
Apache-2.0
def as_top_level_api( logdensity_fn: Callable, step_size: float, inverse_mass_matrix: metrics.MetricTypes, *, max_num_doublings: int = 10, divergence_threshold: int = 1000, integrator: Callable = integrators.velocity_verlet, ) -> SamplingAlgorithm: """Implements the (basic) user interfac...
Implements the (basic) user interface for the nuts kernel. Examples -------- A new NUTS kernel can be initialized and used with the following code: .. code:: nuts = blackjax.nuts(logdensity_fn, step_size, inverse_mass_matrix) state = nuts.init(position) new_state, info = nuts...
as_top_level_api
python
blackjax-devs/blackjax
blackjax/mcmc/nuts.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/nuts.py
Apache-2.0
def iterative_nuts_proposal( integrator: Callable, kinetic_energy: metrics.KineticEnergy, uturn_check_fn: metrics.CheckTurning, max_num_expansions: int = 10, divergence_threshold: float = 1000, ) -> Callable: """Iterative NUTS proposal. Parameters ---------- integrator Sympl...
Iterative NUTS proposal. Parameters ---------- integrator Symplectic integrator used to build the trajectory step by step. kinetic_energy Function that computes the kinetic energy. uturn_check_fn: Function that determines whether the trajectory is turning on itself (...
iterative_nuts_proposal
python
blackjax-devs/blackjax
blackjax/mcmc/nuts.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/nuts.py
Apache-2.0
def init( position: ArrayLikeTree, logdensity_fn: Callable, period: int ) -> PeriodicOrbitalState: """Create a periodic orbital state from a position. Parameters ---------- position the current values of the random variables whose posterior we want to sample from. Can be anything fr...
Create a periodic orbital state from a position. Parameters ---------- position the current values of the random variables whose posterior we want to sample from. Can be anything from a list, a (named) tuple or a dict of arrays. The arrays can either be Numpy or JAX arrays. logd...
init
python
blackjax-devs/blackjax
blackjax/mcmc/periodic_orbital.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/periodic_orbital.py
Apache-2.0
def build_kernel( bijection: Callable = integrators.velocity_verlet, ): """Build a Periodic Orbital kernel :cite:p:`neklyudov2022orbital`. Parameters ---------- bijection transformation used to build the orbit (given a step size). Returns ------- A kernel that takes a rng_key a...
Build a Periodic Orbital kernel :cite:p:`neklyudov2022orbital`. Parameters ---------- bijection transformation used to build the orbit (given a step size). Returns ------- A kernel that takes a rng_key and a Pytree that contains the current state of the chain and that returns a new...
build_kernel
python
blackjax-devs/blackjax
blackjax/mcmc/periodic_orbital.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/periodic_orbital.py
Apache-2.0
def kernel( rng_key: PRNGKey, state: PeriodicOrbitalState, logdensity_fn: Callable, step_size: float, inverse_mass_matrix: Array, period: int, ) -> tuple[PeriodicOrbitalState, PeriodicOrbitalInfo]: """Generate a new orbit with the Periodic Orbital kernel. ...
Generate a new orbit with the Periodic Orbital kernel. Choose a step from the orbit with probability proportional to its weights. Then shift the direction (or alternatively sample a new direction randomly), in order to make the algorithm irreversible, and compute a new orbit from the se...
kernel
python
blackjax-devs/blackjax
blackjax/mcmc/periodic_orbital.py
https://github.com/blackjax-devs/blackjax/blob/master/blackjax/mcmc/periodic_orbital.py
Apache-2.0