I'm trying to wrap my head around some of the more advanced/esoteric search commands. It seems like there's a lot of power there if you know how to harness it (i.e. you're familiar with statistics, probability, and data mining techniques). So, seeing as I'm very much a lay person, and the documentation is a little light sometimes, I'm hoping that someone can educate us all about this command, what it does exactly, and cases where it would be useful in the real-world. My ultimate hope is to post further questions like this about... well, a lot of the search commands, in order to augment the docs a bit and make us all more powerful splunkers. So, are you using this command and, if so, for what?
Here's a related post: Question about analyzefields search command
AF helps you determine how accurately each field predicts the specified field. As an example, look at the following data:
02/03/2011 01:00,st=CA,state_no=1,vote_no=1
02/03/2011 14:00,st=CA,state_no=1,vote_no=1
02/03/2011 01:00,st=MA,state_no=2,vote_no=2
02/03/2011 02:00,st=MA,state_no=2,vote_no=2
02/03/2011 07:00,st=MO,state_no=4,vote_no=1
02/03/2011 08:00,st=MO,state_no=4,vote_no=1
If you run the following search:
* | af classfield=vote_no
You can see that there is a 100% chance (1.0) that my state (state_no) will predict my vote (vote_num), by looking at the accuracy field (acc). You can also see that state is always declared for a vote (cocur = 1).
The use case here is to determine if we can use the data to predict which state will vote for which candidate and with what accuracy we might make a prediction. This is too small a dataset to make accurate predictions, but given a much more representative dataset, I could, with reasonable confidence, predict that a CA or MO voter will pick candidate #1.
HTH
ron
If you have any searches which utilize this command, please chime in and let us know what it's doing for you.
AF helps you determine how accurately each field predicts the specified field. As an example, look at the following data:
02/03/2011 01:00,st=CA,state_no=1,vote_no=1
02/03/2011 14:00,st=CA,state_no=1,vote_no=1
02/03/2011 01:00,st=MA,state_no=2,vote_no=2
02/03/2011 02:00,st=MA,state_no=2,vote_no=2
02/03/2011 07:00,st=MO,state_no=4,vote_no=1
02/03/2011 08:00,st=MO,state_no=4,vote_no=1
If you run the following search:
* | af classfield=vote_no
You can see that there is a 100% chance (1.0) that my state (state_no) will predict my vote (vote_num), by looking at the accuracy field (acc). You can also see that state is always declared for a vote (cocur = 1).
The use case here is to determine if we can use the data to predict which state will vote for which candidate and with what accuracy we might make a prediction. This is too small a dataset to make accurate predictions, but given a much more representative dataset, I could, with reasonable confidence, predict that a CA or MO voter will pick candidate #1.
HTH
ron
Thanks Ron. Good stuff! I'm going to post some more of these, so please keep your eyes peeled and chime in if you can.