Ctrl+K

blackjax.sgmcmc.sgnht

blackjax.sgmcmc.sgnht#

Public API for the Stochastic gradient Nosé-Hoover Thermostat kernel.

Module Contents#

Classes#

SGNHTState

State of the SGNHT algorithm.

Functions#

init(→ SGNHTState)

build_kernel(→ Callable)

Stochastic gradient Nosé-Hoover Thermostat (SGNHT) algorithm.

as_top_level_api(→ blackjax.base.SamplingAlgorithm)

Implements the (basic) user interface for the SGNHT kernel.
class SGNHTState[source]#

State of the SGNHT algorithm.

Parameters:

  • position – Current position in the sample space.

  • momentum – Current momentum in the sample space.

  • xi – Scalar thermostat controlling kinetic energy.

position: blackjax.types.ArrayTree[source]#
momentum: blackjax.types.ArrayTree[source]#
xi: float[source]#
init(position: blackjax.types.ArrayLikeTree, rng_key: blackjax.types.PRNGKey, xi: float) → SGNHTState[source]#
build_kernel(alpha: float = 0.01, beta: float = 0) → Callable[source]#

Stochastic gradient Nosé-Hoover Thermostat (SGNHT) algorithm.

as_top_level_api(grad_estimator: Callable, alpha: float = 0.01, beta: float = 0.0) → blackjax.base.SamplingAlgorithm[source]#

Implements the (basic) user interface for the SGNHT kernel.

The general sgnht kernel (blackjax.sgmcmc.sgnht.build_kernel(), alias blackjax.sgnht.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.

Example

To initialize a SGNHT kernel one needs to specify a schedule function, which returns a step size at each sampling step, and a gradient estimator function. Here for a constant step size, and data_size data samples:

grad_estimator = blackjax.sgmcmc.gradients.grad_estimator(logprior_fn, loglikelihood_fn, data_size)

We can now initialize the sgnht kernel and the state.

sgnht = blackjax.sgnht(grad_estimator)
state = sgnht.init(rng_key, position)

Assuming we have an iterator batches that yields batches of data we can perform one step:

step_size = 1e-3
minibatch = next(batches)
new_state = sgnht.step(rng_key, state, minibatch, step_size)

Kernels are not jit-compiled by default so you will need to do it manually:

step = jax.jit(sgnht.step)
new_state = step(rng_key, state, minibatch, step_size)
Parameters:

grad_estimator – A function that takes a position, a batch of data and returns an estimation of the gradient of the log-density at this position.

Return type:

A SamplingAlgorithm.

Contents