Skip to main content

Advertisement

Table 2 Performance of three sampling algorithms in calculating the marginal likelihood of an analytically tractable example

From: BCM: toolkit for Bayesian analysis of Computational Models using samplers

Dimensions Log marginal likelihood Likelihood evaluations (x1000)
Analytical FOPTMC SMC MultiNest FOPTMC SMC MultiNest
2 −1.75 −1.80 ± 0.68 −1.74 ± 0.39 −1.73 ± 0.29 147 79 18
5 −5.67 −5.98 ± 1.65 −5.66 ± 0.47 −5.73 ± 0.38 287 281 28
10 −14.59 −14.92 ± 3.34 −14.64 ± 0.62 −14.13 ± 0.63 969 521 95
30 −60.13 −61.11 ± 9.10 −59.85 ± 0.97 * 6420 1511 *
100 −255.62 −257.7 ± 24.8 −255.8 ± 1.54 * 96,251 4271 *
  1. The following algorithms were used: FOPTMC feedback-optimized parallel-tempered Markov Chain Monte Carlo [12], SMC automated-temperature sequential Monte Carlo but without ABC approximation [15], and MultiNest [5]. The column ‘Analytical’ gives the marginal likelihood value calculated analytically. (*) indicates that the computation time exceeded the maximal time of 1 h; the other calculations required at most 5 min