Comment by taeric

Meanwhile, far too many shops still use csv. Hard not to see a new entry as slowing the move.

I am curious on the random access benefits. Seems most ML workloads will naturally scan the data sets fairly linearly. Does this maintain parity on that?