You’re exactly right Subutai. I’m running a large swarm again with the max properly set. Will share the results as they come.
Nick > On Apr 16, 2015, at 11:28 PM, Subutai Ahmad <[email protected]> wrote: > > > Looks like the score dropping to 0 coincided with a sharp rise in the data. I > suspect the encoder params (such as min/max) were not set correctly. If the > input hits the max value a constant signal will be sent to the HTM and the > anomaly score will be constant. > > --Subutai > > On Thu, Apr 16, 2015 at 8:38 AM, Matthew Taylor <[email protected] > <mailto:[email protected]>> wrote: > Right after the apnea, the anomaly score dropped to 0 and flatlined. That > doesn't seem right at all. Is your code online anywhere? > > --------- > Matt Taylor > OS Community Flag-Bearer > Numenta > > On Thu, Apr 16, 2015 at 7:00 AM, Nicholas Mitri <[email protected] > <mailto:[email protected]>> wrote: > Hey all, > > I’ve procured data from a CPAP machine (waveform plotted below wrapping from > subplot to the next). > Is this data suitable for HTM anomaly detection and specifically for NAB? > > The peaks in the data are sighs while short flat regions are apneas. The > baseline for breathing also changes as seen by the slower trends in the data. > I’ve tried using HTM and the anomalies I got weren’t very consistent for > sighs (2nd figure). Even worse, apneas were learned immediately, probably due > to being quantized to a single value. > > I’m defending my thesis in 2 weeks and this is the last section to highlight > anomaly detection in HTM. I’d really like the thesis to end on a good note > with one of the better features of HTM so any advice on how to handle this > data is greatly appreciated. > > Thanks, > Nick > <patient1_cpap_25hz.png> > <Screen Shot 2014-12-17 at 12.54.48 AM.png> > >
