I went over the entry diagram and the documentation again. Let me know if this 
is the proper workflow:

Using included detectors:
1- Format new dataset appropriately and add it to a folder under data/.
2- Add ground truth labels to combined_labels.json under the dataset name. (Do 
I need to also include the original user labeled data like AG_labels_v1.json 
under labels/raw/ or will NAB skip the combining algorithm?)
3- use ‘python run.py’ to run the detectors and get score and results files. 

Now for the DUT, I don’t have it written as a class like the inbuilt detectors 
so I wanted to use its anomaly output using path III. 
Will that work? I can’t figure out how to have NAB run on that file exclusively 
since run.py requires a detector specified. 

Also, I tried using a result file with the format included in appendix F 
(timestamp, value, label) but that produced an error when running ‘python 
run.py —score’ that was only resolved when I retained the format of the result 
files already in the folder (including scores and raw anomaly fields).

Thanks of the help Alex!

Nick 




> On Apr 18, 2015, at 6:02 PM, Alex Lavin <[email protected]> wrote:
> 
> Yes this should work, as long as you follow the datafile naming conventions 
> in NAB, the results files are as specified in the whitepaper [1], and the 
> labels are in the correct format of the current json file [2].
> 
> Cheers,
> Alex
> 
> [1] https://github.com/numenta/NAB/wiki#nab-whitepaper 
> <https://github.com/numenta/NAB/wiki#nab-whitepaper>
> [2] https://github.com/numenta/NAB/blob/master/labels/combined_labels.json 
> <https://github.com/numenta/NAB/blob/master/labels/combined_labels.json>
> 
> Alexander Lavin
> Software Engineer
> Numenta

Reply via email to