Aggregating results across models


I was going through Dr.Subutai’s video on multiple fields and one of the questions that he answered how to aggregate results from multiple htm anomaly models. He mentioned that a summation of log anomaly_likelyhood is a good way of doing this. I was wondering if anyone has done this?

Lets say I have two models and each of their anomaly_likelyhood is 0.999. so if do a summation of log 0.999 to the base 10, i get -0.00043 + -0.00043 = -0.00086. so is the idea that we set a threshold for this result (-0.00086) and then take a call to say if there is a system wide anomaly?

1 Like

I think it could be simpler than that. Say you have 40 models giving back a log anomaly likelihood. You might set the summation threshold to 10.0, which would indicate that approx 10 models or 25% of your input is indicating anomaly. Of course you’d need to tune this number to your use-case.