Hacker Read

sli · 2020-08-27 01:49:06+00:00

DNT just became another fingerprinting datapoint, so it ended up being better to just set it to whatever the critical mass is using (which is typically off).

waivej | karma 366 | avg karma 1.63 · | 2017-06-29 14:26:04

I had to switch it to sensitive mode for the same reason. It's also more useful as a relative metric than specific number.

pabs3 | karma 43824 | avg karma 6.39 · | 2021-10-29 22:03:56

They probably reduce the precision to give some plausible deniability in the face of people who have their network connections monitored.

nabla9 | karma 38285 | avg karma 4.67 · | 2017-07-13 14:46:37+00:00

Cern does lots of data analysis where measurement precision is not essential at least in evaluation phase. Switching into higher precision may be needed only for the final calculations It's also possible to scale and normalize data to lower precision.

carllerche | karma 1590 | avg karma 14.45 · | 2021-12-17 13:18:11

Also, thanks for the thought re: digit precision. I am tracking it here: https://github.com/tokio-rs/console/issues/224

yoklov | karma 1767 | avg karma 2.38 · | 2019-04-10 11:14:47+00:00

There are precision issues with -1...1. 0...1 is actually the right choice IMO.

bigbillheck | karma 1619 | avg karma 1.04 · | 2022-09-04 13:41:22

I've stopped using it as I found it far too sensitive to small differences.

fennecfoxy | karma 812 | avg karma 0.74 · | 2023-06-07 08:56:04

Ah fair, I should've known. I suppose the precision is still required for scientific purposes. Thankfully ML stuff now gets more appropriate precision for a speed increase.

dozzie | karma 1448 | avg karma 0.5 · | 2016-02-12 00:01:31

It's still better than using centiinches per millihours.

usaar333 | karma 3846 | avg karma 2.28 · | 2023-06-27 20:43:56

Dropping numerical precision seems to have done that to some degree?

dang | karma 18142 | avg karma 0.25 · | 2014-06-30 19:37:18

Empirically, it doesn't seem to work very well to try for that level of precision.

Kubuxu | karma 971 | avg karma 4.69 · | 2016-11-08 11:19:34+00:00

Maybe it is using single precision.

nobu-mori | karma 37 | avg karma 2.85 · | 2023-02-19 09:55:34

That's great! All we need to do is negate the output and it will be more accurate.

colechristensen | karma 26001 | avg karma 3.7 · | 2022-08-12 13:02:32

You might be thinking of Selective Availability (which limited accuracy on purpose) being turned off in 2000.

guitarbill | karma 4239 | avg karma 2.71 · | 2018-12-01 21:05:58+00:00

Because it's hard to dose correctly, and has a very slim error margin?

syllogism | karma 3356 | avg karma 5.2 · | 2017-10-09 11:04:37+00:00

Does the ONNX thing actually work?

I'm super skeptical of these interchanges, because it seems very difficult to avoid train/test skew. Any difference in detail between the two implementations is a potential problem. I can imagine different order of operations putting out some values by 0.1%, causing 1% eventual loss of accuracy.

reply

hinkley | karma 39933 | avg karma 2.46 · | 2022-11-11 21:55:59

Something like that. The point was accuracy, not throughput.

monocasa | karma 27236 | avg karma 2.94 · | 2021-01-23 18:21:40

Right, but it can sync arbitrary ranges sooner, which is also awful for consistency.

chrischen | karma 11394 | avg karma 2.09 · | 2010-04-11 02:03:46

More precision requires more data.

scyzoryk_xyz | karma 2630 | avg karma 3.91 · | 2021-12-23 01:20:27

My understanding is that nuclear doesn’t do variability very well