Table 1. Simulation parameters.

Parameter	Description
N	Number of sampled chromosomes
S	Number of segregating sites
θ	Population mutation rate for the investigated region(4Neμ)
ρ	Population recombination rate for the investigated region(4Ner)
α	(For selection samples) = 4Nes, where s is the selective coefficient
τ	(For selection samples) time when the beneficial mutation get fixed
t0	(For bottleneck samples) time when the bottleneck ended
t1	(For bottleneck samples) duration of bottleneck

N_e, the effective population size; μ, mutation rate per generation per individual within the investigated region; r, recombination rate per generation per individual within the investigated region. The time parameters (τ, t₀, t₁) are in the unit of 4N_e generations, backward in time. For the selection samples, a beneficial mutation has been simulated to occur at the center of the region. The beneficial allele frequency increases subsequently and fixation was assumed to occur at time τ. The whole region thus experienced a full, hard selective sweep. With boosting, for each value in a set of parameters, 100 replications were simulated as candidates for training, and 100 independent replications were simulated for testing (some testing samples were also used with LDhat, SSL, and PAC). One example scenario involved n = 200 with S = 59 segregating sites. Under this scenario, we simulated samples for different values of ρ. With boosting, we trained for instance under the values ρ = 20, 60, 100, 140, and 180 and estimated ρ under testing samples with values of ρ in {10, 20,…,150}. This leads to 5 × 100 = 500 training samples and 15 × 100 = 1500 testing samples (generated independently from the training samples). We used either θ or S to set the level of polymorphism, and mostly we used fixed S in our simulations.