# Copyright 2019 Artem Artemev @awav, Eric Hambro @condnsdmatters # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import Callable, Sequence, Optional import tensorflow as tf from gpflow.base import Parameter __all__ = ["SamplingHelper"] class SamplingHelper: """ This helper makes it easy to read from variables being set with a prior and writes values back to the same variables. Example: model = ... # Create a GPflow model hmc_helper = SamplingHelper(model.log_posterior_density, model.trainable_parameters) target_log_prob_fn = hmc_helper.target_log_prob_fn current_state = hmc_helper.current_state hmc = tfp.mcmc.HamiltonianMonteCarlo(target_log_prob_fn=target_log_prob_fn, ...) adaptive_hmc = tfp.mcmc.SimpleStepSizeAdaptation(hmc, ...) @tf.function def run_chain_fn(): return mcmc.sample_chain( num_samples, num_burnin_steps, current_state, kernel=adaptive_hmc) hmc_samples = run_chain_fn() parameter_samples = hmc_helper.convert_to_constrained_values(hmc_samples) """ def __init__( self, target_log_prob_fn: Callable[[], tf.Tensor], parameters: Sequence[Parameter] ): """ :param target_log_prob_fn: a callable which returns the log-density of the model under the target distribution; needs to implicitly depend on the `parameters`. E.g. `model.log_posterior_density`. :param parameters: List of :class:`gpflow.Parameter` used as a state of the Markov chain. E.g. `model.trainable_parameters` Note that each parameter must have been given a prior. """ if not all(isinstance(p, Parameter) and p.prior is not None for p in parameters): raise ValueError( "`parameters` should only contain gpflow.Parameter objects with priors" ) self._parameters = parameters self._target_log_prob_fn = target_log_prob_fn self._variables = [p.unconstrained_variable for p in parameters] @property def current_state(self): """Return the current state of the unconstrained variables, used in HMC.""" return self._variables @property def target_log_prob_fn(self): """ The target log probability, adjusted to allow for optimisation to occur on the tracked unconstrained underlying variables. """ variables_list = self.current_state @tf.custom_gradient def _target_log_prob_fn_closure(*variables): for v_old, v_new in zip(variables_list, variables): v_old.assign(v_new) with tf.GradientTape(watch_accessed_variables=False) as tape: tape.watch(variables_list) log_prob = self._target_log_prob_fn() # Now need to correct for the fact that the prob fn is evaluated on the # constrained space while we wish to evaluate it in the unconstrained space for param in self._parameters: if param.transform is not None: x = param.unconstrained_variable log_det_jacobian = param.transform.forward_log_det_jacobian( x, x.shape.ndims ) log_prob += tf.reduce_sum(log_det_jacobian) @tf.function def grad_fn(dy, variables: Optional[tf.Tensor] = None): grad = tape.gradient(log_prob, variables_list) return grad, [None] * len(variables) return log_prob, grad_fn return _target_log_prob_fn_closure def convert_to_constrained_values(self, hmc_samples): """ Converts list of unconstrained values in `hmc_samples` to constrained versions. Each value in the list corresponds to an entry in parameters passed to the constructor; for parameters that have a transform, the constrained representation is returned. """ values = [] for hmc_value, param in zip(hmc_samples, self._parameters): if param.transform is not None: value = param.transform.forward(hmc_value) else: value = hmc_value values.append(value) return values