test results

["eberhard speer jr." <se...@ducis.net>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi,

ran a number of tests using a number of test data sets.

Cutting to the chase : using data collected over a period of a couple
of weeks from a 'known' server's web-access logs, 31,457 unique
ua-strings resulted in 53 'unknowns' [see attached]

I did however use the 1.0 'new' data set 'augmented' with the 'old'
ODDR data otherwise all Nokia Series... come tumbling out as 'unknown'
pushing up the total, in this data set to 103

Using larger test data sets [sub-sets of the one in the repository]
and comparing 'clean' 1.27 ODDR and to 'clean' 1.0 DeviceMap, as
expected, the results are much better and adding the older device
data, the 'unknowns' over a set of almost a million user-agent string
drops to 7,000 [all mangled strings, real odd ball or junk string]

Things that jump out when looking at larger test results are :
Haier         market/vendor specific [regional]
Huawei
Lenovo        laptops...an issue
LGE           older models ?
MOT-....      older models !
Nokia         older models

So now tests depend on the source of the data and/or what is being test...

But, as expected : tweaking the builder patterns for n-grams did
improve the results by a big margin.

If I notice 'important' stuff I'll obviously report them, otherwise
feel free ask for more details.

esjr
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.8 (MingW32)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJTuY6bAAoJEOxywXcFLKYcWHoH/ik7y2BJSCVXIRNdrXU/74sV
7qbPXY9dKBKtTy0tgQpCM10lpxpM9ma1qwgKaFcTO+MuHTGNAS/t6T/q4vUYnHMi
Wr3Lwy0lHWns4GWal1jrzfM96+ikLvUff+iCEL9A/AtCSFiNFOGD9LTNlsLxlw2b
qbaBkHgrAmdzXS5OH/JJ1+iOIVL6QkqRXshZXsw1VvL81nY7395ikIoHTopIt4PJ
UPUqd7Jgz1Vuzo+M0Ywl1d/gsK7cUnwn6vymJlpXhjOKCdyZRkcXw8fm3Nx6LziV
RHbWlTUqx+AzBJ9c2n+sDasgGakxRMHQg905D05zJr4ZkliZS8A4Bl5U9+Guj8Y=
=5WCm
-----END PGP SIGNATURE-----

Re: test results

["reza.naghibi@yahoo.com.INVALID" <re...@yahoo.com.INVALID>]

Just an FYI... Regarding finding new and unknown devices, you should remove the generic fallback patterns from the builder patch file first. This will expose the unknown devices more clearly, otherwise, they can be swept up as generic devices. They are mainly the patterns on the bottom in the simple builder.