FDE: Do we have any numbers on performance?Hi,
I have a semiconductor customer who have trace lines for 10000 parts on 25 wafers per lot. Each trace line has about 1000-2000 frequency steps measured. The curves look similar to these: (though this image is from a paper Don and Hadley wrote)
(the binary files with the data are about 500MB-2GB large, so could be huge in RAM).
Thanks,
Martin
Re: FDE: Do we have any numbers on performance?Thanks Mike, I haven't thought of this option, but yes that's worth a try.
However, I think I have also to set expectations right that this will take a while no matter what machine they will use. I will also check on what the data is based on, may one could find candidate sets or cluster to reduce complexity and dimensionality.
Re: FDE: Do we have any numbers on performance?Setting aside the issue of needing a ton of memory for a second, that sounds like case for the Direct PCA option in FDE.
But it's going to take a bit to run regardless. What you're describing is almost like a hyperspectral data set and those take a long time to run - even on purpose built systems with GPU acceleration.
M
