Fig. 6.

Comebacks of actors are unpredictable. a Typical profile sequence of an actor exhibiting a comeback after a long period of latency. Such cases might lead to misclassification when the subcareer fed to the prediction algorithm (highlighted in pink) captures a long latency period: the prediction algorithm wrongly classifies the pink sequence as one where the annus mirabilis has passed. b Semi-log probability distribution of the estimated time lapse from the (wrongly estimated) annus mirabilis to the true one (i.e., the time t_cb to come-back for actors with profile sequences such as the one in panel a), for those misclassified samples where the algorithm wrongly predicts that the annus mirabilis had already passed (a linear binning has been applied to the data). Modelling the position of the secondary peak (comeback burst) as a random variable, the fact that P(t_cb) decays exponentially, suggests that this random variable is memoryless (Poisson process), i.e., the comeback burst is intrinsically unpredictable