PGINN: trained networks and Drosophila Adh posterior genealogies

Trained neural networks and migrate-n output supporting the preprint "Population-Genetics-Informed Neural Networks for a Single-Population Coalescent Under Selection" (P. Beerli). Code: https://github.com/pbeerli/PGINN pginn-models.tar (1.0 GB): all 234 networks (PyTorch .pt) and feature scalers (.npz) trained by code/reproduce.sh: MSE-only and PGINN-loss networks for every sample-size/locus regime and 10 training replicates, the w sweep, the w=1 check and the neutral-limit check. File names match those that reproduce.sh expects in code/model/. pginn-adh-migrate.tar (99 MB): migrate-n 6.1.11 posterior genealogy sample (trees.tre), outfile, infile and parmfile for the 15 four-gamete segments of the Drosophila melanogaster Adh region in 176 DGRP lines (Jukes-Cantor model, random seed 20260924, 16 MPI processes). Unpacks into code/data/adh-migrate/. SHA256SUMS: checksums of both archives. Running ./fetch_data.sh in the code repository downloads, verifies and unpacks both archives. With them, every figure and table of the paper can be rebuilt without retraining or rerunning migrate-n.