Alex, 

There seems to be a discrepancy between the scoring function in the 
documentation and in the code. The figure in the documentation has the window 
being limited by relative positions -3 and 0.
The scorer.py file sets the left edge as -1. It also assigned a value of 0.88 
to the center while the documentation figure has a flat value of 1 from far 
left to center. 

Should I assume this part of the documentation obsolete and go with the code 
instead?

thanks,
Nick

> On Apr 22, 2015, at 6:34 PM, Alex Lavin <[email protected]> wrote:
> 
> Hi Nick,
> Comparing raw scores vs the normalized scores is really a matter of 
> preference; the normalized scores, with a set range of 0-100, may be easier 
> to comprehend.
> 
> Precision, recall, and their harmonic mean, F1 score, are helpful evaluation 
> functions for most machine learning algorithms. They don't, however, 
> incorporate time in the calculations, and are thus unsuitable for evaluating 
> the ability of an algorithm to perform on real-time, streaming data. A main 
> motivation for NAB was to design a scoring system which incorporates time and 
> the TP,TN,FN, and FP counts. I wouldn't call this problematic, but rather 
> desirable that NAB scoring doesn't mimic other evaluation functions.
> 
> Cheers,
> Alex
> 
> Alexander Lavin
> Software Engineer
> Numenta

Reply via email to