sampling_builder#
- deepinv.sampling.sampling_builder(iterator, data_fidelity, prior, params_algo={}, max_iter=100, thresh_conv=1e-3, burnin_ratio=0.2, thinning=10, history_size=5, verbose=False, callback=lambda X, **kwargs: ..., **kwargs)[source]#
Helper function for building an instance of the
deepinv.sampling.BaseSampling
class.See Uncertainty quantification with PnP-ULA. and Markov Chain Monte Carlo for example usage.
See the docs for
deepinv.sampling.BaseSampling
for further examples and information.- Parameters:
iterator (SamplingIterator | str) – Either a SamplingIterator instance or a string naming the iterator class
data_fidelity (DataFidelity) – Negative log-likelihood function
prior (Prior) – Negative log-prior
params_algo (dict) – Dictionary containing the parameters for the algorithm
max_iter (int) – Number of Monte Carlo iterations
burnin_ratio (float) – Percentage of iterations for burn-in
thinning (int) – Integer to thin the Monte Carlo samples
history_size (int | bool) – Number of most recent samples to store in memory. If
True
, all samples are stored. IfFalse
, no samples are stored. If an integer, it specifies the number of most recent samples to store. Default: 5verbose (bool) – Whether to print progress
callback (Callable) – A function that is called on every (thinned) sample state dictionary for diagnostics. It is called with the current sample
X
, the currentstatistics
(a list of Welford objects), and the current iteration numberiter
as keyword arguments.kwargs – Additional keyword arguments passed to the iterator constructor when a string is provided as the iterator parameter
- Returns:
Configured BaseSampling instance in eval mode
- Return type: