Figure 6
From: Efficient discovery of responses of proteins to compounds using active learning
Accuracy of models constructed using sampling by active learning approaches. A CCT model was trained using all experiments chosen by each selection method up to a given round. The mean and standard error of the area under a ROC curve for each selection method is plotted versus the fraction of experiments performed up to that round. The methods were random choice (red), greedy CCT (green), greedy CCT-uncertainty hybrid (blue), and greedy selection-uncertainty hybrid using memory limits of five (cyan) and ten rounds (magenta).