NAB scoring returns minus score for each true negative

Hi there, I am implementing NAB with as final goal to start testing with HTM.

Right now, I am just trying to implement the checked out version of april this year of the NAB github. I started running it with the random detector to gain an understanding of the whole project. However, when I check the results after the scoring step, for each row which is a true negative I get a minus score of the amount of the false positive weight. So for example for the standard profile each true negative is scored with -0.11, since the fpWeight = 0.11.

(I was testing with the datasets of realKnownCause)

When I compare this to the results in the version of github, it looks very different, there the true negatives just get a score of 0. I really don’t understand what is going wrong because I didnt change any code.

I hope someone can point me to the right direction of what I am overlooking.

Any help is appreciated!

This seems related to:

You are the author of this ticket, correct? I will address the github issue first. Thank you for reporting, this, @michellea.

Hi, yes thats correct! Today i found the issue so my forum question is kind of outdated now. Thanks!


Let’s continue this discussion on the Github issue.