Distribution of classification probabilities in different stages of the learning loop. The top panel shows classification results from the classifier, trained with ten alerts (five SNIa and five randomly selected from other classes). This initial state was allowed to evolve through 300 iterations following different query strategies. Bottom left: results from a training sample built following a random sampling strategy. Bottom right: results from a training sample built following an uncertainty sampling strategy. The vertical dotted line in each panel denotes a 50% probability.