Fig. 1.

image

Download original image

Illustration of conditional (pure Gibbs; orange) and marginal sampling (blue) algorithms for a highly correlated (Pearson’s correlation coefficient of ρ = 0.99) two-dimensional Gaussian distribution (black contours). The initial position (βinit, ainit) is indicated by a purple dot. Left: comparison of un-normalized conditional P(β ∣ ainit) distribution evaluated at the initial position and the corresponding marginal P(β) distribution; note that the latter is much wider than the former. Assuming β samples drawn at the 10th percentile, the graphs along the vertical lines represent un-normalized conditial distributions P(a ∣ β) evaluated at the β values drawn with the conditional (orange) and marginal (blue) distributions of β; note that the marginal sampling case results in a much longer step length between the initial and final sample values. Right: samples of a standard Gibbs sampling chain (orange) using conditional sampling for both a and β, and a sampling chain (blue) using marginal sampling for β and conditional sampling for a. Both cases show the first 100 samples initialized from the purple point.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.