dfm / emcee / 29206614742

96%

push

github

fix: draw an independent proposal per walker for full-covariance GaussianMove (#571)

rng.multivariate_normal was called without a size argument when a
full covariance matrix was supplied, so a single random vector was
generated and broadcast to every walker's position. This made all
proposed steps identical within an iteration, breaking the ensemble
walkers' independence for GaussianMove(cov) with a 2D cov matrix.

Signed-off-by: Mike German <mike@stepsventures.com>

279 of 324 branches covered (86.11%)

13 of 13 new or added lines in 1 file covered. (100.0%)

1638 of 1702 relevant lines covered (96.24%)

3.44 hits per line