Skip to main content
Figure 10 | Journal of Cheminformatics

Figure 10

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

Figure 10

Estimating the uncertainty profile for ensemble models constructed using the averaging method. Thin and thick blue lines represent the distributions of predictions for logP3-3a and logP3-3b, respectively; the corresponding error and error rate distributions are shown in red and green. Uncertainty profiles and fitted beta binomials for error distributions are represented by dashed lines. Fitted prediction distributions have been omitted for clarity. Dotted vertical lines indicate thresholds for summed outputs and do not separate negative predictions and errors from positive ones (see Figure 9 and text). (A) Distribution of errors and predictions for the training pool. (B) Distribution of uncertainties and observed error rates for the training pool. (C) Distribution of errors and predictions for the validation set. (D) Distribution of uncertainties and observed error rates for the validation set.

Back to article page