Darnell Granberry

Training on large datasets raises OOM error before training. Need some alternative such as storing in normal RAM or loading on-the-fly.

Samplers in topaz/utils/data/sampler.py need adjusting to keep track of individual sizes instead of group[0]'s only