On Sun, Aug 4, 2024 at 11:57 PM Undescribed Horrific Abuse, One Victim & Survivor of Many <gmkarl@gmail.com> wrote:
it sounds like the challenge maybe could simplify to the idea of, i have some numbers that are multiplying many other numbers, and some of them come out to really low values. i want to not have to load the numbers that will come out to really low values.
maybe karl would consider for example grouping them based on what is in the data? for example you could index them based on what high values multiplying them make high value outputs. this sounds like the kind of idea that could actually work :s :s :s
[so it would of course depend on the detail of the module in question, but you could for example have a run-encoded list of each input, showing which rows or columns respond to it, and you'd only have to store the indices or lengths of each run, as a single vector of indices. the size of this might be the square root of the original size divided by the bitdepth ratio such as 4 [--