Samplers¶

class geoopt.samplers.RHMC(params, epsilon=0.001, n_steps=1)[source]¶

Riemannian Hamiltonian Monte-Carlo.

Parameters:	params (iterable) – iterables of tensors for which to perform sampling epsilon (float) – step size n_steps (int) – number of leapfrog steps

step(closure)[source]¶

Perform a single sampling step.

Parameters:	closure (callable) – A closure that reevaluates the model and returns the log probability.

class geoopt.samplers.RSGLD(params, epsilon=0.001)[source]¶

Riemannian Stochastic Gradient Langevin Dynamics.

Parameters:	params (iterable) – iterables of tensors for which to perform sampling epsilon (float) – step size

step(closure)[source]¶

Perform a single sampling step.

Parameters:	closure (callable) – A closure that reevaluates the model and returns the log probability.

class geoopt.samplers.SGRHMC(params, epsilon=0.001, n_steps=1, alpha=0.1)[source]¶

Stochastic Gradient Riemannian Hamiltonian Monte-Carlo.

Parameters:	params (iterable) – iterables of tensors for which to perform sampling epsilon (float) – step size n_steps (int) – number of leapfrog steps alpha (float) – \((1 - alpha)\) – momentum term

step(closure)[source]¶

Perform a single sampling step.

Parameters:	closure (callable) – A closure that reevaluates the model and returns the log probability.

geoopt