Skip to main content
Fig. 6 | BMC Bioinformatics

Fig. 6

From: Significance evaluation in factor graphs

Fig. 6

a We investigate the quality of the approximation as a function of graph size and conditioned on the degree of independence between variables as measured by the variance ratio (18). Here we found the 1%-quantile in a Markov chain using importance sampling. We then found the saddlepoint approximation of the tail probability in this particular point and plot the relative error as a function of the length of the Markov chain. For details see Additional file 1: Section ii. b The relative error measured at different quantiles for both SA and normal approximation, this was done under the same range of conditions as above. We see that while the errors remain stable for the SA they increase for the normal approximation as we move to smaller percentiles

Back to article page