diff --git a/docs/dynesty-guide.txt b/docs/dynesty-guide.txt index f4d98c0dfb306c75debe3e901d809174744b11b5..b04a6d40d4f64877a8bf7df7fd5a2a2df86b8159 100644 --- a/docs/dynesty-guide.txt +++ b/docs/dynesty-guide.txt @@ -38,8 +38,9 @@ new point from the constrained prior. These can be specified using the results, however, in some cases, a larger value may be required. 2. :code:`sample="acceptance-walk"`: with this method, at each iteration all MCMC chains - are set to the same length. The specific length evolves during the run to yield an - average of :code:`naccept` accepted jumps during each chain. This method is well + are set to the same length. The specific length evolves during the run so that the + number of accepted steps follows a Poisson distribution with mean :code:`naccept` + during each chain. This method is well suited to parallelised applications, as each MCMC chain will run for the same amount of time. The value of :code:`naccept` should be tuned based on the application. For example, one could run a single analysis with @@ -62,7 +63,9 @@ There are a number of keyword arguments that influence these sampling methods: for consistency with other codes. Default is 5000. If this limit is reached, a warning will be printed during sampling. -#. :code:`proposals`: a list of the desired proposals. The allowed values are +#. :code:`proposals`: a list of the desired proposals. + Each of these proposals can be used with any of the sampling methods described + above. The allowed values are * :code:`diff`: `ter Braak + (2006) <https://doi.org/10.1007/s11222-006-8769-1>`_ differential evolution. This is the default for :code:`bound="live"` and @@ -70,7 +73,9 @@ There are a number of keyword arguments that influence these sampling methods: * :code:`volumetric`: sample from an ellipsoid centered on the current point. This is the proposal distribution implemented in :code:`dynesty` and the - default for all other :code:`bound` options. + default for all other :code:`bound` options. This was the default proposal + distribution for :code:`Bilby<2`, however, in many applications it leads to + longer autocorrelation times and struggles to explore multi-modal distributions. Finally, we implement two custom :code:`dynesty.sampler.Sampler` classes to facilitate the differential evolution proposal and average acceptance tracking.