search

mathiesonlab / pg-gan

33 stars 9 forks source link

question about training #26

Open mufernando opened 3 months ago

mufernando commented 3 months ago

Hi,

Thanks for the great piece of software!

It is not clear to me from a somewhat quick glance at the codebase whether the training is done on an entire chromosome (250Mb) or if the windows are treated as iid (5000 windows each 50Kb)?

Thanks,

Murillo

saramathieson commented 3 months ago

Windows (50kb regions) are treated as iid. During training, windows are chosen randomly throughout the genome to create each batch. It wouldn't be too difficult to experiment with one chromosome at a time if you're interested though. Thanks!