The Poco sampler¶
Preconditioned Monte Carlo with normalizing flows
Name |
poco |
Version |
0.1.1 |
Author(s) |
Minas Karamanis, David Nabergoj, Florian Beutler, John Peacock, Uros Seljak |
URL |
|
Citation(s) |
https://arxiv.org/abs/2207.05660, https://arxiv.org/abs/2207.05652 |
Parallelism |
parallel |
pocoMC relies on the Preconditioned Monte Carlo algorithm which utilises a Normalising Flow in order to decorrelate the parameters of the posterior, and facilitates efficient sampling of probability distributions with non-trivial geometries.
It generates both posterior samples and Bayesian evidences.
PocoMC does not currently support checkpointing (all the output appears at the end of the run) or derived parameters.
Installation¶
if using conda, do this first to ensure consistent dependencies:
conda install -c conda-forge corner pytorch tqdm
pip install pocomc
Parameters¶
These parameters can be set in the sampler’s section in the ini parameter file. If no default is specified then the parameter is required. A listing of “(empty)” means a blank string is the default.
Name |
Type |
Description |
Default |
|---|---|---|---|
n_particles |
integer |
Number of particles/walkers |
|
nsample_add |
integer |
Number of extra posterior samples to draw at the end |
0 |
threshold |
float |
The threshold value for the (normalised) proposal scale parameter below which normalising flow preconditioning (NFP) is enabled. Default is to always enable. |
1.0 |
scale |
bool |
Whether to scale the distribution of particles to have zero mean and unit variance |
N |
rescale |
bool |
Whether to rescale the distribution of particles to have zero mean and unit variance in every iteration |
N |
diagonal |
bool |
Use a diagonal covariance matrix when rescaling instead of a full covariance |
T |
ess |
float |
The effective sample size maintained during the run |
0.95 |
gamma |
float |
Threshold for the correlation coefficient that is used to adaptively determine the number of MCMC steps |
0.75 |
n_min |
int |
Minimum number of MCMC steps per iteration |
ndim/2 |
n_max |
int |
Maximum number of MCMC steps per iteration |
ndim*10 |
seed |
int |
A random seed for reproducibility. Default is to generate automatically. |
0 |