APBRmetrics

A better simulation that understands more about how the game works at factor level? I'd probably begin with a cluster analysis of players by broad position (PG, wing, big) into 3 "types" and 3 quality levels using box score stats / metrics, player tracking data and RPM. 27 player clusters, 4x2 factors. That seems doable, worthwhile starting parameters for model. With time go to something like 50-80 player clusters and maybe 16-64 sub-factors (capturing a lot of the player tracking data).