Mario, sorry it took so long to respond. It's up to you to decide where the threshold will be for the anomaly likelihood score where you decide it's actually anomalous. From the chart you showed, it doesn't look like it gets up to 1.0 very often, especially after it sees more data. In that chart, how many rows of data has it seen? Do the scores get lower if you run it longer?
--------- Matt Taylor OS Community Flag-Bearer Numenta On Thu, Oct 9, 2014 at 2:22 AM, Mario Tambos <[email protected]> wrote: > Hi everyone, > > I'm trying to fit a CLA model for anomaly detection on ECG data, and I > wanted to as whether eny of you has any experience with this. > So far my results are far from optimal. > I have 15 minutes worth of ECG with a sample every .004 seconds. > I swarm over the first 1000 samples and then start feeding the model > with the complete ECG from the start. > The predictions look good, but the model seems to be assigning to each > QRS complex an anomaly_likelihood=1. You can see an excerpt of the > data here: > https://gist.github.com/mtambos/70159bdf59dcfe300b57 > > and a plot here: > https://www.copy.com/s/nXTViyi2fxOMdIMZ/cla_ecg.png > > Here's also the model params generated by the swarm > https://gist.github.com/mtambos/269298118704f5bb39a8 > > and here the swarm description > https://gist.github.com/mtambos/5f642555945da8ddb484 > > Any help/tips would be greatly appreciated. > Thanks! > > Mario > >
