Skip to main content
Figure 7 | Journal of Cheminformatics

Figure 7

From: Using beta binomials to estimate classification uncertainty for ensemble models

Figure 7

Effect of increasing the number of networks in an ensemble. Models logP3-2a and logP3-2b were comprised of 75 networks rather than of the default 33. Symbols have been omitted for clarity. (A) The validation set error distribution for logP3-2a (for which the naïve threshold of 37.5 was used) is shown in red and the error distribution for logP3-2b (refined voting threshold: 53.5) is shown in green. The corresponding thresholds are shown as vertical dotted lines and the fitted beta binomials are shown as dashed lines. The distribution of predictions and its fitted beta binomial are the same for both models; they are represented by the solid and dotted blue line, respectively. (B) The uncertainty profile and observed error rates for models logP3-1a and logP3-1b are shown in red and green, respectively. Observed and calculated uncertainties are represented by solid lines and dashed lines.

Back to article page