How fast is nupic?

I’m working on a project that I think would benefit from real-time anomaly detection, but the computation ultimately needs to be done on a phone, so I’m curious if anyone has any intuitions on feasibility. The data will be about 10 signals streaming at about 20 samples per second. Thoughts?

1 Like

it depends on which platform do you use? c++ or Python?
I use C++, so that for my test prediction framework (scalar encoder + TM + decoder) I reach 1000Hz and for L4L2Column for prediction around 100Hz at my Ubuntu 16.04LTS.

1 Like

It is not going to be easy getting NuPIC running on a phone.

Write your own! :smiley:

Or if it’s an Android phone, maybe it’s worth taking a look at

1 Like

I doubt Android would tolerate HTM.Java’s threading model… I think there are strict guidelines for the way Android treats those?

I’ve done a lot of work on Android. It’s got the same threading API as regular JVM java. I don’t see why it wouldn’t work, but you can try it and see if it passes the tests.