Fig. 4

Structure of the GP4PG algorithm. From an initial set of models, we compute a summary statistic (4jSFS) for each simulation and compare it to the observed data by the means of a fitness error function. Then, the errors are ranked. We produce the offspring (2nd generation) following an invasive weed optimization algorithm modifying each child model (mutation). We repeat this procedure until the error is 0 or it reaches a plateau