Fig. 3

Stability of the predictions for subsets of observations. This figure summarizes the stability of the predictions for all samplings of the observations, compared to the final predictions with all 100% of observations. “Good” predictions (matching the 100% predictions) are depicted in green, “Bad” predictions (predicted differently than the 100% predictions) in red and “Missing” predictions (not predicted) in blue. For each category, four curves are plotted representing, from top to bottom, the maximum, median, mean and minimum number of predictions of this type. Curves are normalized to the number of predictions obtained for each set of sampled data