a few insights i've had recently:
- multi-dimensional dataframes are like keywords are like hyperedges; the multi-dimensional-ness is like multiple arguments to a function are like multiple roles to a hyperedge; names in dataframes (like dict keys) are like keywords are like role names; accessing dimensions out-of-order (slices) are like partial function application are like currying the hyperedge; the distinction between completely-specified slices and individual values is like the distinction between a 0-ary function and its output, is like the distinction between a path and its endpoint.
- present-day GPUs and OpenCL? may be a sufficient framework to simulate Connection Machine-style Active Data Paradigm, since they can do upwards of 64k-way data-parallelism, even though only ~8-64-way synchronizable-shared-memory parallelism. So, target OpenCL?
![[Home]](http://www.bayleshanks.com/cartoonbayle.png)