3 ms·
" cannot use this data to recreate anything remotely resembling the original dataset." This by itself may be mostly true perhaps - and many of the comments get
by stevenicr 9y ago
" cannot use this data to recreate anything remotely resembling the original dataset."
This by itself may be mostly true perhaps - and many of the comments get into ways of playing with this dataset to make it better, I don't have experience with those methods, but,
what I have not seen anyone mention, if you have this dumbed down dataset, the original is gone.. you can still combine with other data sets that are either public or previously created and likely fine tune;
dumbed down set + public voter records + public arrest records + previous whatever records - sort, match, what's left over.
and pretty much recreate what you needed from the original, maybe not 100%, but I would guess you could get really close.
- firebones 9y agoOr "better" (for some value of better), join the data to something identifiable (e.g., the public records you listed) before developing models for everything you want to retain, then discard the original data once your deep enough net has effectively auto-encoded whatever you wanted to retain in an identifiable manner.