search

snd297 / braintrust

Automatically exported from code.google.com/p/braintrust
0 stars 0 forks source link

mismatch between filtered data segments #282

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago 
downloading an interval of filtered data gives a different result than 
downloading two filtered subsegments of data and joining them.

Original issue reported on code.google.com by benbrinkmann on 31 May 2012 at 7:11

Attachments:

GoogleCodeExporter commented 9 years ago

Original comment by john.fro...@gmail.com on 1 Jun 2012 at 7:51

GoogleCodeExporter commented 9 years ago 
Hi guys, is there a standard for figuring out how much we need to pad each 
window?  I assume it would be based on native + target sampling rates, etc. If 
we can figure out what the right scheme is it should be pretty easy to 
implement.

Original comment by zack.i...@gmail.com on 3 Jun 2012 at 6:56

GoogleCodeExporter commented 9 years ago

Original comment by zack.i...@gmail.com on 6 Jun 2012 at 3:52

GoogleCodeExporter commented 9 years ago

Original comment by samd.p...@gmail.com on 10 Aug 2012 at 6:09

GoogleCodeExporter commented 9 years ago

Original comment by samd.p...@gmail.com on 10 Aug 2012 at 6:15

GoogleCodeExporter commented 9 years ago

Original comment by john.fro...@gmail.com on 14 Sep 2012 at 8:22

GoogleCodeExporter commented 9 years ago 
Sam, support for padding should be in place for decimation.

Original comment by zack.i...@gmail.com on 19 Oct 2012 at 12:03

GoogleCodeExporter commented 9 years ago

Original comment by samd.p...@gmail.com on 19 Oct 2012 at 5:29

GoogleCodeExporter commented 9 years ago

Original comment by samd.p...@gmail.com on 25 Oct 2012 at 7:12

GoogleCodeExporter commented 9 years ago 
This one may not be done yet. I wrote a test (w/ no asserts)
edu.upenn.cis.db.mefview.server.TestFilterManager.segmentsSameAsAll(). Section 
A of the test just decimates the first page of samplePages, section B just 
decimates page 2, section B decimates all of samplePages. But the results for 
the second page are different for B and C, but they should be the same if I 
understand correctly.

There should be 150 samples of padding for all calls to 
FilterManager.decimate(...) in the test.

I'm using filter type 0. The results are also different when using filter type 
FilterSpec.NO_FILTER

Original comment by samd.p...@gmail.com on 26 Oct 2012 at 7:14

GoogleCodeExporter commented 9 years ago 
Note that between rounding error and linear interpolation, there's no guarantee 
that these are going to line up in a way that produces 100% identical results.  
Is this more significant than would be expected there?

Original comment by zack.i...@gmail.com on 26 Oct 2012 at 7:22

GoogleCodeExporter commented 9 years ago 
Does it make sense for the results to not be equal when there is no filter 
being applied and when downsampling by an integer factor?

Here's the 1st 4 values of page 2 after running Decimate on just samplePages[1]:
1035,1069,1079,1089,

Here's the 1st 4 values of page 2 after running Decimate on all of samplePages:
1060,1072,1082,1092,

The raw values are:

1060,1070,1080,1090

That 1035 looks suspect to me - is it?

Original comment by samd.p...@gmail.com on 26 Oct 2012 at 7:56

GoogleCodeExporter commented 9 years ago 
It looks like that 1035 is the last value on samplePages[0]. So it should not 
end up as the first value of the processed samplePages[1]. 

Assuming my code is correct, I think there is a problem, especially since 
samplePages[0] and samplePages[1] are separated by a gap.

Original comment by samd.p...@gmail.com on 31 Oct 2012 at 6:34

GoogleCodeExporter commented 9 years ago

Original comment by samd.p...@gmail.com on 31 Oct 2012 at 6:40

GoogleCodeExporter commented 9 years ago 
Even if the gap is eliminated in samplePages, the same problem happens.

Original comment by samd.p...@gmail.com on 31 Oct 2012 at 6:42

GoogleCodeExporter commented 9 years ago

Original comment by samd.p...@gmail.com on 7 Aug 2013 at 6:21